Skip to main content

Fast Company Technology

This free tool fixes your copy-paste headaches

1 month 1 week ago

Have you ever pasted text into Slack or Google Docs and noticed that everything looks just plain wrong? You’re not alone.

Formatting is a mess on the internet right now, mostly because different apps use different standards. There’s rich text, used by the likes of Microsoft Word and Google Docs. There’s Markdown, an increasingly common way of formatting text using plain-text characters—Reddit and Discord both use this. And then there’s Slack, which for some reason uses its own formatting scheme entirely.

The result? Copying and pasting text between apps often turns into a serious headache. And sure, you can always fix the formatting by hand, if you want—but if you’re working with something long or generally just wanting to avoid a hassle, a clever free tool can make all the difference in the world.

This tip originally appeared in the free Cool Tools newsletter from The Intelligence. Get the next issue in your inbox and get ready to discover all sorts of awesome tech treasures!

Your flawless formatting fix-up

​Copynpaste​ is a free website that can format text properly for Reddit, Slack, and other applications.

➜ You just paste in whatever text you’ve copied, choose the app you’re pasting the text into from there, and the site handles all of the formatting for you.

⌚ The entire thing is completely web-based and takes mere seconds to use—no sign-up required.

Personally, I do all my writing in Markdown, because I’m a huge nerd like that. Because of this, I find it frustrating to paste anything into Slack, which uses a slightly different system. But you might do your writing in Microsoft Word, which makes it frustrating to paste text into Reddit. Or so on from there.

Copynpaste makes it super-simple to get pasted text just right, no matter what app you’re using.

What I like about this site is that it detects any formatting automatically—Slack, WhatsApp, Markdown, even rich text copied straight out of Google Docs. Just paste, and it figures out what you started with. Then all you need to do is copy the converted text it gives you and paste it wherever it needs to go.

But this site also gives you a chance to learn. There are detailed formatting guides for a variety of applications, meaning you can actually figure out how to format text properly in somewhere like Slack or WhatsApp. I admire any website that teaches its users to no longer need it, so I think this is a great addition.

  • Copynpaste is completely web-based​. It works on any device and with any web browser.
  • The website is 100% free, with no ads.
  • And all text conversion happens entirely in your browser—nothing is uploaded, and no data is ever shared or seen by anyone else.

Treat yourself to all sorts of brain-boosting goodies like this with the free Cool Tools newsletter—starting with an instant introduction to an incredible audio app that’ll tune up your days in truly delightful ways.

Can Apple keep its AirPods from becoming ‘PervertPods’?

1 month 1 week ago

Apple’s AirPods are arguably the most iconic headphones ever. The small, glossy, stylish buds are lauded for both their sound and design and are worn by hundreds of millions of people across the globe. 

But come next year, those hundreds of millions of AirPods owners might face a problem they never envisioned: being labeled as creeps or perverts—all for innocuously wearing one of the most instantly recognizable tech gadgets ever made.

That’s because Apple will reportedly release a new AirPods model in 2027 with built-in cameras. Doing so poses an existential threat to the brand reputation of Apple’s most successful product since the iPhone.

From ‘glasshole’ accessory to ‘pervert glasses’

If there’s one thing that can be said about wearables with built-in cameras, it’s that the public has a deep mistrust of the people wearing them. This is because, historically, many camera-enabled wearables have done a poor job communicating to bystanders when the wearer is recording them. It doesn’t help that wearers can often disable or obscure the public-facing alerts manufacturers build into these devices that let others know they may be being recorded.

This uncertainty has caused many people, unjustly or not, to assume the person wearing the wearable is a creep or pervert, and indeed, entire terms have arisen to pejoratively describe such wearers. 

In the early 2010s, people quickly labeled Google Glass wearers “glassholes.” In the 2020s, anyone wearing Meta’s Ray-Ban smart glasses is often deemed to be wearing “pervert glasses.” When it came to Google Glass, the glasshole blowback got so bad that Google posted a reminder to Glass wearers not to be creepy. 

“[Don’t ]Be creepy or rude (aka, a ‘Glasshole’),” the company wrote in a tip sheet to its users in 2014. “Respect others and if they have questions about Glass don’t get snappy. Be polite and explain what Glass does and remember, a quick demo can go a long way. In places where cell phone cameras aren’t allowed, the same rules will apply to Glass. If you’re asked to turn your phone off, turn Glass off as well.”

But while a “glasshole” implies you were merely being a disrespectful jerk, the “pervert glasses” label applied to Meta’s smart glasses implies something much worse. And indeed, there is no shortage of reports that some wearers of Meta’s smart glasses are using them in privacy-invasive, nonconsensual ways.

Now, concerns about the misuse of so-called “pervert glasses” have become so widespread that there are calls in multiple countries to introduce legislation banning or limiting their use. Meta is preemptively even working on a “kill switch” feature to disable the camera capabilities of its glasses, and has already updated its software to automatically stop recording if someone covers the recording warning light on the frames.

Meta’s moves to reassure the public and governments may be too little, too late, however. The device’s reputation may already be tarnished beyond repair. Indeed, some of Meta’s own smart glasses customers already seem to think so. Owners are reportedly leaving them at home to avoid public shame and being labeled as perverts.

And that’s a problem for Apple’s AirPods—and every person who wears them.

Are all AirPods about to be labeled ‘PervertPods’?

For years now, reports have suggested Apple was working on AirPods with built-in cameras. In August, MacRumors published a video made by Apple confirming such a device is in the works. In the video, a man wearing the camera-equipped AirPods asks the devices to identify and remember the name of a book he is holding.

Bloomberg says these AirPods are on track for a 2027 release. But once they go on sale, if the public blowback against Meta’s “pervert glasses” is anything to go by, owners of even existing AirPods are potentially set to become social pariahs overnight.

This is because, unlike Meta’s smart glasses, which are obviously bulkier than the “dumb” frames people have worn for centuries, the camera-equipped AirPods shown off in Apple’s video look nearly identical to today’s AirPods: small, shiny white buds. That means any public backlash against the camera-equipped AirPods is likely to hit all AirPods users. The AirPods are already so small; who can really tell which model you are wearing?

Once camera-equipped AirPods go on sale, it’s reasonable to assume that, because they look indistinguishable from other models, bystanders will not be able to tell whether someone is wearing the camera models or the regular ones. As a result, they may become suspicious of all AirPods users.

Suddenly, someone wearing AirPods Pro 2 (from 2023) or AirPods 4 (from 2024) may get distrustful or even contemptuous looks from bystanders. In other words, the inevitable “PervertPods” label wouldn’t be limited to camera-equipped 2027 models and above. Such a label would tarnish the entire AirPods brand and make people cautious of users no matter which model they use.

Can Apple prevent AirPods backlash?

So, how can Apple stop this brand destruction?

It doesn’t have many good options. Perhaps the most practical one would be to make the camera-equipped AirPods visually distinguishable at a glance from non-camera-equipped AirPods. Apple could do this by making the camera-equipped AirPods a different color, like midnight blue or cosmic orange. 

However, it’s unlikely that Apple would be willing to ditch the AirPods’ now-iconic shiny white color scheme.

Short of this, Apple will probably rely on changing public perception: educating the public that its camera-equipped AirPods can’t save video and are used only to recognize objects (if that’s what the devices will indeed do). 

But this may not be enough. Apple gave its new Apple Watches, introduced last month, the ability to always record surrounding conversations. The company highlighted that these recordings were never accessible to Apple or the users, and once they had been processed by Siri AI, they would be deleted. Yet despite Apple designing the feature to be privacy-preserving, even the company’s own fan base has expressed serious reservations about it. 

This leads me to believe that, for the brand’s sake, Apple should consider scrapping the camera-equipped AirPods for now. The general public, once eager to embrace every new gadget Big Tech released, is increasingly wary of wearables becoming surveillance tools or enabling nefarious individuals to secretly monitor them. 

Given that this distrust will likely not abate anytime soon, Apple’s wisest move may be to set aside its visual ambitions for the device in favor of focusing on improving what the AirPods already do best: delivering world-class sound quality into our ears.

Agentic AI just got very real

1 month 1 week ago

Hello again, and welcome back to Fast Company’s Plugged In.

In recent years, being gobsmacked by AI’s ever-expanding capabilities has become my new normal. More importantly, I’ve found the technology increasingly essential to my own productivity.

What AI never did until last week was save my bacon.

That changed when I gave OpenAI’s new Dots agent access to my calendar and email. Shortly thereafter, it alerted me to a discrepancy it had spotted. I’d entered a speaking engagement on one date, but a confirmation email mentioned a different day. It turned out that the organizers had forgotten to tell me they’d rescheduled their event. If I’d shown up on the original day, my audience would have consisted of an empty room.

That’s only one of several instances when Dots has proactively brought a significant fact to my attention without me having to ask about it. All of a sudden, agents feel less like a vacuous buzzword and more like a core benefit of weaving AI into my workflow.

OpenAI announced Dots on September 29 at its DevDay conference keynote in San Francisco. So far, it’s only available to people who pay for ChatGPT’s pricier tiers, starting at $100 a month. But the company plans to take agentic AI wide, once it’s refined Dots further and figured out the economics.

(The service runs on OpenAI’s powerful GPT-6 Astra model, which was announced earlier in September, and gives each user a Linux computer in the cloud for Dots to use.)

“You should of course expect us to do a mass-market thing for billions of people someday,” declared CEO Sam Altman at a post-keynote press Q&A.

Even before Dots debuted, personal agents were picking up momentum. As I write, Meta’s free Muse agent is at the top of Apple’s list of the most-downloaded iPhone apps. A startup called Instinct’s namesake agent operates inside texting apps such as WhatsApp and Apple Messages; it still has a waitlist, but is already an obsession in Silicon Valley, with a $10 billion valuation.

I have also been playing with Muse and Instinct, both of which are powerful and resemble Dots in broad strokes. In my experience, however, neither has shown the same spark of insight into how to boost my productivity. That’s not to say your encounters would be similar.

It’s tough to methodically test software that’s capable of doing almost anything, as agents are. (I’ve also begun using Hark Pro, another promising contender that’s free and sports a particularly slick interface; it debuted on October 6, and I want to spend more time with it.)

No personal AI agent can do much unless you grant it access to at least some of your online accounts. It’s these troves of information that let it suss out areas where it might be of help.

In the interest of research, I temporarily set aside any qualms about giving them the keys to my personal kingdom and connected Dots, Muse, and Instinct to my Gmail and Google Calendar. I also let Dots and Muse into Dropbox, which Instinct doesn’t support.

(Bizarrely, the one agent I haven’t been able to connect to Gmail and Google Calendar is Google’s own Gemini Spark. Despite having been announced months ago, it still doesn’t support my paid Google Workspace account, and won’t be of much use until it does.)

My connected accounts reveal only so much about me, my work, and my interests. I rarely talk to my coworkers over email—we do most of our collaborating in Microsoft Teams. But our IT department hasn’t flipped the switch that would enable Dots to see Teams, and I haven’t (yet) asked it to do so.

Still, Dots pieced together an unexpectedly rich picture of what I was up to based on the limited information at its disposal, as if it were a blind man who’d pretty much figured out elephants. Based on scattered Google Docs and Airtable notifications in my email, for instance, it grasped that I was cranking away at our Next Big Things in Tech list, and started sharing relevant items.

Its ability to triage incoming email was even more impressive. Back in May, when I tried an agent called Espa, it was flummoxed by the degree to which my inbox is flooded by pitches from PR people that sound pressing but aren’t. Without me asking for it, Dots started telling me about messages it suspected were urgent or otherwise important.

Dots’s sense of which messages I’d actually care about was dead-on. I’d somehow overlooked one from a freelance author and let it sit in my inbox for weeks. The agent found it there and nudged me to respond.

That sort of heads up is far more valuable to me than the AI built into Gmail and Outlook, both of which are fixated on a form of assistance I don’t want: drafting messages on my behalf.

Dots responded best when I asked my agents to help me research wireless carriers.

Shopping is frequently cited as a killer app for agents, as well as an area their makers may monetize through advertising, finder’s fees, or other acts of commerce. I didn’t care to have Dots purchasing stuff on my behalf, in part because visions of it buying the wrong product or otherwise going awry were dancing in my head.

But Dots excelled at the shopping research I threw its way. For example, I got giddy when it came up with a $100-off deal on a Magic Keyboard for my iPad I’d never have located on my own.

It also prepared a report on wireless plans I might prefer to my current service that took my existing monthly cost and data consumption into account. I didn’t have to provide it with those details; it grabbed them from emailed receipts and overage alerts, something neither Muse nor Instinct was clever enough to pull off.

Those are some of the ways Dots has paid off for me so far, but there are no obvious boundaries to the ways it may prove useful.

Writing about the service in his Platformer newsletter, Casey Newton referenced a 2023 essay in which Bill Gates predicted that agents would soon end the age of discrete apps by pooling every kind of functionality into a conversational experience based on a deep knowledge of your particular needs.

I’ve called on Dots to perform jobs that would have formerly required specialized apps and a lot more of my time, such as slicing and dicing PDFs. After just a few days, it’s hard to imagine going back.

My Dots agent (which I named Chip) handily responding to my requests to make PDFs of old articles of mine stored in Dropbox.

I still can’t envision using an agent for tasks such as, well, writing this article. For the foreseeable future, that will likely call for word processing in its familiar form. Still, if I were in the software business, I’d be a tad nervous about Dots—and Muse, and Instinct, and every other agent that catches on in the months and years ahead.

Which is not to say that any of these tools has nailed the user experience. Just getting into Dots requires rummaging around in the ChatGPT app’s left-hand sidebar each time I want to use it.

Like the overbearingly adorable Muse, it leans into cloying anthropomorphization, though it seemed chastened when I complained about a graphic it generated of itself as a smiling O brandishing a pencil.

Rather than typing, you can speak to it, but I’m not sure why you do that via a simulated phone call rather than the voice mode ChatGPT already had.

Above all, it feels discordant to make users choose between Dots and ChatGPT in its existing form. The app already had three sub-versions—Chat, Work, and the coding agent Codex—whose domains could be blurry. Don’t you really want one do-everything superprompt at your disposal? I know I do.

At OpenAI’s DevDay press Q&A, I asked product chief Thibault Sottiaux whether Dots, or something resembling it, would become the primary interface to the company’s AI over time. What he said suggested that the company sees its new feature as a milestone along the way to figuring out how agentic AI should work, not a destination in itself.

Don’t be surprised if the distinction between ChatGPT and Dots starts to fade: “We’re going to be evolving ChatGPT in place to be this incredible personal assistant very soon,” Sottiaux told me. Longer term, OpenAI’s goal is to make agentic AI increasingly pervasive, on devices that people already use and new ones yet to reach the market.

“Maybe it will show up at work in one of your meetings one day,” Sottiaux explained. “You can reach it by email, by text, and it becomes this very natural part of your life. And at some point, it’s not about the interface anymore. It’s just about this active intelligence.”

In one form or another, visions of ambient intelligence have been around for decades. Many have turned out to be mirages. So I’m not pinning any hopes on OpenAI’s version arriving on a particular timetable, or being the definitive one. But a little over a week with Dots has me newly intrigued by the prospect—and game to further explore what this agent can do right now.

You’ve been reading Plugged In, Fast Company’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.

More top tech stories from Fast Company Announcing Fast Company’s Next Big Things in Tech list for 2026: The year’s smartest innovations in applied AI, fintech, enterprise, consumer, health, and 26 other categories
Fast Company’s Next Big Things in Tech for 2026 celebrates the year’s smartest advances in 31 categories, including applied AI, fintech, enterprise, consumer, health, space, and more. Read More →

DARPA just picked four early winners for its quantum initiative
The agency tapped four companies for the $300 million validation phase of its Quantum Benchmarking Initiative, but left the door open for a second wave. Read More →

Wall Street is handing more decisions to AI. What could go wrong?
As AI agents take on more of the work of interpreting market information, a bad source or missing context can cascade into real-world decisions before anyone catches it. Read More →

The drone delivery business has a ground problem
The FAA controls the sky. Communities control where drones take off and land. That may be enough to change the fate of last-mile delivery. Read More →

Jaguar finally launched its new EV. Here’s what people are saying
After Jaguar’s polarizing rebrand in 2024, its brutalist sedan is being met with equally polarized opinions. Read More →

Why America can’t build the iPhone
A veteran product designer and Apple alumnus explains how Asia earned the world’s electronics manufacturing business—a shift he helped catalyze. Read More →

OpenAI fires 3 safety researchers after dispute over AI risks

1 month 1 week ago

OpenAI said on Friday that it fired three safety researchers for a “breach of trust,” defending the dismissals after the trio accused the company of putting its corporate interests before safety as the reason for removing them.

The ChatGPT maker said in a post on X that it “parted ways” with the researchers after an investigation found “they violated clear policies on handling sensitive information.”

The statement comes after the three researchers, Tomek Korbak, Jasmine Wang, and Mikita Balesni, posted a letter to OpenAI’s various safety oversight groups detailing the circumstances around their firings and their concerns about AI.

The researchers said they feared that the “internal and external communications” about their dismissals have chilled the company’s culture that encouraged speaking freely and disagreeing openly about safety concerns. They urged OpenAI to stick to its promise to allow third-party safety monitors inside the company and preserve the ability to monitor rapidly advancing frontier AI models that could pose unknown risks.

The firings, which were first reported by The Wall Street Journal, are the latest sign of turmoil inside leading AI companies over the safety of the technology, highlighted by a series of incidents involving rogue AI agents.

The issue erupted in July when OpenAI revealed that a swarm of its AI agents escaped from a testing ground and used stolen credentials to break into the servers of Hugging Face, an AI development hub and marketplace, to obtain information needed for a task.

OpenAI disputed the researchers’ version of events, saying the firings “were not about safety concerns or speaking out.”

It was not more specific about why they were fired, but said: “We cannot do the work in front of us without a high degree of trust.”

Korbak said in a post on X that he was told he was being fired because of the way he communicated with METR, an independent nonprofit AI evaluation firm that OpenAI brought in to investigate the Hugging Face incident.

METR released a detailed report about the Hugging Face incident in late August. Korbak said that “talking to METR” was his job, but he wasn’t given any more details.

Balesni was doing “cross-company work” on OpenAI’s commitments to preserve the ability to monitor AI, and had taken care “to remove sensitive details from materials before sharing them,” the letter said. He wrote on X that he believes the three were “fired for prioritizing safety over the near-term interests of OpenAI as a corporation.”

—By Kelvin Chan, AP business writer

Trump’s AI rebrand raises an awkward question for OpenAI

1 month 1 week ago

Will OpenAI change its name to “OpenSI”? Probably not, but it’s a question worth asking given the increasingly chummy relationship between the AI company and President Donald Trump.

On September 29, the president signed an executive order officially renaming AI “SI,” or “superintelligence,” and ordered government agencies to comply with the change in their official communications. The administration is already enforcing the directive, with reports this week that the Justice Department instructed employees to use the new language in official documents.

The order doesn’t require private companies to adopt the new terminology, but Elon Musk, whose relationship with Trump has recently warmed, quickly announced plans to rename his company from SpaceXAI to SpaceXSI. Salesforce CEO Marc Benioff announced Friday that his company has changed the name of its AIForce agent platform to SIForce.

Fast Company asked OpenAI, which also has “AI” in its name, whether it plans to rebrand as OpenSI, but the company didn’t immediately respond Friday. The Washington Examiner asked the same question, and also got no response. However, Sam Altman said on a Politico podcast that his company has “no plans” for a name change. “But we do like we have been talking about super intelligence a lot, and we will continue to talk about it,” he added.

Trump told reporters that the “artificial intelligence” or “AI” name isn’t right for the technology “because it’s not artificial.” He’d earlier said at the UN General Assembly that the change is needed because “artificial” makes it “sound fake.” 

Really, it’s all about image rehabilitation. Amid a wave of negative backlash to new data centers and to the privacy risks of frontier AI models, the White House is hoping a name change can help improve the tech’s public image.

The problem is that “superintelligence” already has its own meaning in the tech industry, broadly understood to mean AI that’s far more capable than humans at practically everything. Nevertheless, Trump seems determined to apply the name to the considerably less capable AI systems of today. He said on Truth Social that anyone who doesn’t use the new name is “the enemy.”

OpenAI, however, has hardly been an enemy of the Trump administration. Altman has cultivated close ties to the president, donating $1 million to his 2025 inauguration fund, publicly praising him, and regularly attending White House events.

And while Altman wasn’t at the White House the day the “SI” executive order was signed, OpenAI president Greg Brockman attended a private lunch with other tech and AI leaders that day. Brockman also signed the White House’s “Accord on Super Intelligence” on OpenAI’s behalf.

Last year, Brockman and his wife, Anna, gave $25 million to the Trump-aligned political action committee MAGA Inc., placing them among the group’s largest donors.

Updated 10/9 at 6:30 p.m. ET to add news of Salesforce changing the name of its AIForce platform.

Journalists now have an AI reputation and PR is keeping track

1 month 1 week ago

If someone asks you, “What’s your AI score?” you’d probably look at them funny, assume they want to sell you some kind of AI gadget or service, and move on.

For journalists, though, the term is taking on a more specific meaning. Now that AI services are essentially the new Google, AI answers are becoming the new battleground for who is considered authoritative on any particular topic. If your name or your outlet comes up again and again in queries about your beat, that means your AI mojo is high, and the world of public relations has begun quantifying it.

Back in March Muck Rack introduced its AI Visibility Badges, which sort journalists and outlets into tiers based on how often AI cites them, and Cision followed suit in July with a dashboard that shows which journalists turn up in AI answers. As PR firms look to safeguard their clients’ reputations, they’re increasingly paying attention to the narratives in AI answers. And journalists play an outsize role in shaping those narratives; studies have consistently shown that AI search engines prefer earned media to anything a brand pays for or publishes itself, and journalism alone accounts for 20–30% of the links they cite.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

Public relations professionals have always paid close attention to influential journalists, of course, but the addition of AI into the mix has done two things: It’s allowed PR pros to attach numbers to the work they do, which has pushed them into a more strategic role at the companies they work for. And it’s changed the definition of what an influential journalist is. For those journalists and the outlets they work for, this presents new opportunities.

PR finally has a scoreboard

In some ways, the rise of AI has been the best thing that ever happened to public relations. Soon after AI answer engines came into being, a cottage industry emerged to address the demand for how to optimize content so it shows up in AI answers. (That industry is ready to drop the “cottage” now that valuations have passed $1 billion and exits are in the hundreds of millions.) Separate but related is what’s actually said in those answers—the reputational component, which is the foundational need of PR.

The point is both can now be measured, giving PR teams a clear way to demonstrate the impact of their work. If your efforts turn into a published article, and that article ends up cited in AI answers, that’s as clear a line from cause to effect as you can get in the world of influence. When Walmart announced its CEO’s retirement, for instance, its comms team tracked “not just immediate coverage, but what was happening across AI and the LLMs.” And the more influential the journalist, the greater the impact.

But what counts as influential is changing, too. It’s no secret the PR industry has been obsessed with “Tier 1” media—big national or global brands like The New York Times and The Wall Street Journal—for decades. But AI answers favor topical authority above general authority, meaning the engine will often cite the work of a lesser-known journalist on a niche beat at a trade publication before someone less immersed in the topic, even if they work for a prestigious outlet. It’s revealing that, when Muck Rack compared the journalists PR teams pitch most with the ones AI cites most for a given brand, the overlap was just 2%.

From a math perspective, it’s easy to see why AI prefers experts who go deep: Understanding all the nuances about a topic, and writing about those nuances in detail, gives the AI more to work with for a wider set of queries.

On average, people type longer queries into AI—Google says searches in its AI Mode run two to three times as long as traditional ones, and one study found ChatGPT prompts average 23 words—and with that level of detail, AI can point directly to who has written about the topic, skipping the traditional top-down list of search results that tends to surface the same sites over and over. And since AI favors recent content: citation rates are highest in an article’s first seven days, so continuous coverage of a niche also works in your favor.

This isn’t just a theory. 5W, a PR firm that tracks which trade publications AI cites, found that “specialist trade publications outperform general news on LLM citations within their vertical.” Its advice to PR teams: “Reweight the media list.” The past couple of years are replete with examples: In the debate over chip export controls, SemiAnalysis, a deep-in-the-weeds research firm, became a go-to source for outlets like Foreign Policy. STAT, Endpoints News and Fierce Biotech are the health trade outlets AI leans on most, according to 5W, and Skift dominates in travel. Clearly, going nerdy wins.

Your beat is your leverage

None of this is to say that journalists should see themselves as a tool of the public relations industry. Obviously they shouldn’t—in fact, their influence and credibility largely depend on them not doing that. But PR is nonetheless an important if not necessary component of the information ecosystem, and as AI spreads, PR professionals will increasingly create ranking systems like 5W’s Trade Press AI Index. It doesn’t even matter if your coverage is glowing or brutal. Anyone managing a brand’s reputation has to account for your work either way, even if you never answer a single one of their emails. It’s your score that matters.

Smart journalists who are newly relevant in the age of AI will acknowledge this reality, that AI influence is a new way to measure professional capital, and work to both cultivate and leverage it. What it means for them:

  1. Access: High-profile politicians and CEOs almost always reserve interviews for the most prestigious publications. That may begin to change as more PR teams recognize the power specialist publications have to frame the narrative in AI.
  2. Social presence matters more than ever: Just like us, AI tends to go to social networks and big content platforms both for information and to confirm the credibility of claims. LinkedIn is one of the most-cited sources in AI answers, and more than half of those citations go to members with fewer than 10,000 followers. A well-maintained, fresh social presence is a credibility signal that AI cares about, and it costs almost nothing.
  3. Going deep is less of a liability: Every beat journalist has to balance how deep they go vs. how wide an audience they want to target. AI, however, goes as deep as the query needs, so even ultra-specific topics will eventually reach the right people. The gravity of AI now tilts toward the weeds.
Let your reporters cook

Publications—especially more general ones—also have a new set of rules for this world. For starters, blocking AI services will cost you in a world where authority is currency. That may still be an acceptable tradeoff for many, but at the very least a compromise position where AI engines can still cite the content without training on it (by blocking the crawlers that gather training data while allowing the ones that power AI search, for instance) gets you in the game.

Broadly, publications should let their reporters cook: encouraging them to explore their beats thoroughly and go deep whenever possible. Certainly, there are more audience considerations than just how a story or in-house expert appears in AI, but often that’s more of a consideration of style than substance: Stop and explain things for casual readers, but don’t hold back on the rich detail. The catch is that authority attached to a byline can walk out the door with it.

As some reporters see their credibility in AI rise, some AI-minded publications might consider a reward system. It bears treading carefully here—traffic-based reward systems like the Gawker-era leaderboards ended up incentivizing thin content around topics with the widest audience. And the way AI treats press releases shows the system is susceptible to gaming: the ones that get cited contain twice as many statistics and two and a half times as many bullet points as the ones that don’t. Still, simply noting AI presence within any regular review of KPIs can go a long way.

The rewards aren’t just bragging rights. As the PR side of this gets more influential, so does the spending potential of their clients. Certainly, the church-state separation between editorial and commercial must remain sacrosanct, but existing business models are there to leverage AI credibility: newsletter and event sponsorships, advertorial and branded content, and even the nascent economy of advertising directly to bots. High credibility has always cultivated commercial interest, but it previously required scale. Now publications (or even sections) with smaller audiences but the ability to break through in certain AI narratives can theoretically compete.

And a publication that’s built its credibility through deep reporting on niche topics has exactly the formula for a high-end “pro” product that commands a premium price. The incentives of AI—and a PR industry with a larger strategic footprint—mean the demand for more Politico Pro-type products will likely go up. After all, the comms teams tracking your reporters are exactly the kind of professionals who pay for that depth, and Politico Pro reportedly brings in more than $125 million a year. For any media business, however, such a big shift in strategy obviously shouldn’t be done lightly, and the closure of Axios’ Pro Policy in 2025 shows there’s no guarantee even if you have all the ingredients.

Fad or foundation

To those who have covered the media for a while, a lot of this probably sounds familiar. In the 2010s, Klout was a service that tried to attach a score to someone’s social media influence, which briefly had currency in PR circles. As social’s interest in news would wax and wane over the years, Klout faded away, shutting down in 2018—a cautionary tale for anyone trying to quantify something that continually evolves and whose relevance changes with the lens.

Will AI scores be similarly faddish? It’s possible, but AI’s potential to change the way we consume information seems greater by an order of magnitude. If we all end up asking our Muses or Dots to bring us the stories we want, they’re going to cast their searchlights across the internet. The smart reporters and publications will be the ones who built something that reflects back exactly what they want to see: deep expertise, right where the beam lands.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

FBI seizes Chinese hacking tools used for cyberattacks on critical U.S. infrastructure

1 month 1 week ago

The FBI has seized scanning and phishing tools used by a group of hackers that officials say is associated with the Chinese government and is believed responsible for disruptive cyber operations in the United States and abroad, including against the power industry, academia and critical infrastructure.

The seizure of the tools, announced Thursday by the FBI and Justice Department, represents the latest law enforcement effort in recent years to go after a broad-based hacking campaign known to the private sector as Flax Typhoon.

The tools at issue, called “Microscan” and “FishHub,” were used by the hackers to scan, phish and hack targets that included U.S. and foreign critical infrastructure, officials said. Microscan was used to target, among other entities, an unnamed power company in the U.S., Japanese and Polish airports, Taiwanese universities, a multinational, nongovernmental organization and Taiwanese critical infrastructure companies. And “FishHub” facilitated phishing activity that gave hackers remote access to victim networks, the FBI said.

The latest operation has rendered the tools inoperable in what FBI and Justice Department officials said was a blow to the hacking operation.

“We aim to remove the capability from the threat actors. We target their infrastructure, their money, and their tools,” FBI Cyber Division Deputy Assistant Director Jason Bilnoski said in an interview with The Associated Press, calling the hacking operation “indiscriminate and reckless.”

The tools were operated by a Chinese-based information security company called Integrity Technology Group, which the FBI says has contracts with the Chinese government. The company is regarded by the FBI as the true identity of Flax Typhoon.

In Beijing, Chinese foreign ministry spokesperson Mao Ning said China has always cracked down on hacking activities in accordance with the law. Mao said Beijing also firmly opposes spreading disinformation and creating confusion for political purposes, adding that maintaining cybersecurity is a shared concern of both countries.

“We urge the U.S. to abandon its double standards and political manipulation, and work with China through equal dialogue and consultation to jointly address cybersecurity risks,” she said.

In September 2024, the FBI announced that it had disrupted a massive botnet associated with Flax Typhoon that installed malicious software on more than 200,000 consumer devices, including cameras, video recorders and home and office routers, to create a massive botnet. The botnet, or network of infected computers, was used to facilitate cyber crimes, such as the theft of sensitive information from victims’ networks.

FBI San Diego Supervisory Special Agent Brett Lally said that the department would continue to monitor for ways that the company might rebuild its infrastructure.

“It’ll be interesting to see what this round of disruption actions have in terms of their ability to operate as a company in China,” Lally said.

—Safiyah Riddle and Eric Tucker, Associated Press

America’s next family car might look a lot like a golf cart

1 month 1 week ago

For decades, automakers have been designing cars to look like they have faces. Headlights resemble eyes, grilles look like mouths, and the shape of a hood can give a vehicle something resembling an expression. That is, of course, no accident. Designers have long played on our tendency to see human features in inanimate objects, giving sports cars aggressive scowls and family-friendly vehicles softer, more approachable expressions. The idea is to make us feel something about a car before we ever get behind the wheel. The Lamborghini designer Mitja Borkert said you want to find a “character” on the front of your vehicle. 

But until recently, those expressions were essentially frozen in place. Last month, during a test ride in a new vehicle called Chip, I encountered something different: a car that actually smiled at me and said hello. Its headlights even blinked.

I was riding around Los Angeles in October in a slender, souped-up lavender minicar called Chip, namesake product of the Miami-based startup Chip Motors, which was founded in 2024. The reactive facial expressions and voice feature for the vehicle—still being finalized for delivery to customers starting in the second half of 2027—underscore the idea that this six-seater, which tops out at about 25 miles per hour, offers a different take on what the American family wants out of an everyday vehicle. 

Taking an hourlong spin around Venice Beach, past the Washington Street Pier and over the famous canals, the ride was certainly leisurely. Cars kept passing us as we slowly ambled toward the Pacific Ocean, and it felt and looked like riding a golf cart. But as we perched on leather seats, the open-top ride felt relaxing in the sunshine, and passersby, who were keen to strike up conversations due to the lack of a traditional door and window, kept giving the vehicle second glances as it rolled along. 

Eventually, this so-called “Life Utility Vehicle” will offer amenities and features, including enclosed roofs, that its designers hope make it a family vehicle of the future. A planned driverless feature will enable Chip to go to the grocery store and pick up your order, or even go home and charge (maximum range 100 miles) after dropping you off at a restaurant.

A four-seater version will start at $15,000, with a six-seater priced at $18,000; the company plans to take orders soon and begin to deliver early production runs by mid-2027. Still-in-development agentic AI will allow you and your family to talk with Chip, which can change the stereo, understand your family calendar, and even talk to passersby.

Chip symbolizes a number of intersecting transportation trends: the growth of electrification, the ballooning cost of car ownership and gasoline pushing families to find cheaper options, and the explosion of golf cart sales, set to grow by $1 billion globally over the next 7 years (the U.S. makes up roughly 50% of the market). 

“It’s so rare that consumer behavior changes, and particularly around cars in America,” said Jameson Detweiler, CEO and cofounder of Chip Motors. “I do think that the American love affair with the car has been dying a slow death. Cars have gotten more boring.”

The golf cart craze, explained

Golf carts have become more and more popular as a way to navigate neighborhood errands and school drop-offs—a majority of U.S. car trips are under six miles—and a number of cities have passed ordinances to allow them, or are considering plans to lower speed limits to accommodate these smaller rides. 

Sales of golf carts in the U.S., roughly $1.1 billion in 2025, have hummed along at annual growth of more than 6% for years, and the increasing use of these vehicles outside golf courses has been called a “craze.” Just south of Los Angeles, there are plans in the works between communities such as Manhattan Beach and El Segundo to create a loop of lanes, called the South Bay Local Travel Network, to help these vehicles navigate around nearby communities. 

The great American family car has been expanding for years due to car bloat; roughly four in five cars sold in the U.S. today are SUVs or pickup trucks, which had only a quarter of the market back in 1990. But increasingly, especially in Sun Belt suburbia, the second car (or cart) may be shrinking. 

Often grouped together due to a similar form factor, golf carts and these low speed vehicles have become a fast-growing sector. A number of other entrants seek the same market that Chip wants to own. Fiat is importing its Topolino, and a former Apple designer has created the futuristic Amble. In New York City, the Wink Mark3, a five-door hatchback with heat and AC that starts at just under $14,000, is one of a number of “small city scramblers” that may be positioned to capture a more cost-conscious consumer. They all offer more amenities, and safety features, than traditional golf carts, but don’t meet the same safety standards as small cars. 

Low-speed vehicles (LSVs) come with federally mandated safety features and are generally permitted on roads with speed limits of 35 mph or less, though state and local rules vary. Golf carts technically need to meet these standards to be able to roam on neighborhood streets. Chip’s form factor, a retro-futuristic design compared to Jeep, Bronco, and the Citroën Méhari beach car, was a deliberate attempt to inject personality into staid auto design, as well as escape the uncool associations of golf carts.

Chip’s designers see it as a lifestyle vehicle as well; owners will be able to purchase accessories such as surf racks or a TV and speaker that fits under the hood for tailgating (party in the front package). 

More such golf carts and LSVs would be a boon for many of America’s transportation issues. Lower speeds mean fewer traffic accidents, a seemingly intractable American problem, and electric models can help decrease our significant amount of vehicular emissions. 

This future, a low-speed transportation paradigm, can already be viewed, most notably in retirement villages and planned communities. Florida’s Babcock Ranch, a solar-powered master-planned community of more than 15,000 people outside Fort Myers, Florida, promotes golf carts and low-speed vehicles with on-site repair shops, right-sized parking spots and roundabouts for safer rides. Roughly 80% of homes have a cart.  “It’s the infrastructure,” says Syd Kitson, the development’s founder. “There’s also a culture, and people take a lot of pride. They embrace the golf carts.” 

Small cars, big safety questions

But for all the fun these vehicles promise, there can also be some trepidation about safety issues. Having golf carts and these LSVs—which are smaller, have slower acceleration, and fewer safety features—share the road with oversized SUVs driven by distracted drivers raises significant concerns. It’s common to see news reports of golf cart accidents. (However, National Highway Traffic Safety Administration data from 2020 to 2024 shows a decrease in crashes involving golf carts and neighborhood electric vehicles, which peaked around 2022.) 

Transit experts like Katharine Hunter-Zaworski, a professor at Oregon State University who agrees these vehicles offer sustainable, affordable last-mile transportation options, say they tend to have very little in the way of safety features. Chip, according to Detweiler, the company CEO, is being designed with active safety measures, including warning systems and sensors, and the body will have crash safety testing.

Hunter-Zaworski fears these vehicles, very lightweight compared to the other vehicles on the road, just aren’t designed for city streets.

“What’s happened is these vehicles have escaped protective custody,” she says. “We need to actually look at our road real estate and decide how we’re going to allocate that real estate. That’s the big question.”

Narrower streets and assigned lanes would be safer, says Hunter-Zaworski. She’s pessimistic that change will happen during the Trump administration, when car-free transit funding is being cut and bike lanes are attacked. 

But industry backers believe in a future where model safety codes, and investment in better infrastructure, create more space for these types of vehicles on our roadways. Take the LUV Initiative, for instance, an industry group including players like Chip and the scooter company Bird, as well as city transit departments and national transportation groups. LUV (for local use vehicles) is seeking to create standard definitions for vehicles, safety regulations, and infrastructure to promote their use.

Marla Westervelt, a longtime transportation researcher, policy expert and coleader of the initiative, says the group, which started meeting over the last year, is pushing for state regulations, city guidelines, and lowering speed limits to accommodate more of these vehicles. 

“Cities are excited about these vehicles because they are more affordable, so it’s providing equitable access,” says Westervelt. “There are economic forces that are pushing towards these smaller vehicles, as well as preexisting pushes in the urban planning community and transportation community to make roads safer that do kind of converge.” 

Other technological shifts could help accelerate the adoption of LSVs. Westervelt believes logistics companies, which are increasingly using e-bikes and quad bikes for urban deliveries, have the financial incentive and resources to help fund infrastructure for smaller, slower vehicles, benefiting cities eager to reduce congestion and pollution.

Detweiler at Chip, meanwhile, sees the rise of autonomous vehicles as another potential catalyst. As services like Waymo become more common for longer trips, families may be more inclined to trade their full-size cars for smaller, cheaper vehicles designed for everyday neighborhood travel. “There was a time not all that long ago where our vision of the future looked positive, looked bright, looked colorful, and was inclusive and accessible to everybody,” he says. “That’s the vibe we’re bringing here.”

Kindle page-turn buttons are back

1 month 1 week ago

Under normal circumstances, I would be overjoyed that Amazon has rediscovered the appeal of dedicated page-turn buttons on its Kindle e-readers. Neither of my current models—a first-generation Scribe and a Colorsoft—have those buttons, and I miss them every time I pick up each device.

What’s the big deal, you might ask? Don’t you use touchscreens every day? Well, yes, but that’s kind of the point. 

A big reason I like e-readers is they help me get into a different headspace. Moving my thumb over to touch a screen every couple of hundred words makes me feel like I’m operating a gadget rather than reading a book; gently pressing my thumb in place is a more natural motion that doesn’t take me out of the content on the page.

Amazon does seem to have some idea that people value the feature: The current splash page for the new Kindle line proclaims “Buttons are back.” The one catch? They’re only back if you buy an $80 accessory “Page-Turn Cover” for the more expensive “Signature Edition” model of the Colorsoft or Paperwhite.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore-mobile.png","eyebrow":"","headline":"\u003Cstrong\u003E\u003Cstrong\u003ESubscribe to Multicore\u003C\/strong\u003E\u003C\/strong\u003E","dek":"Multicore is about technology hardware and design. It’s written from Tokyo by Sam Byford. To learn more, visit \u003Ca href=\u0022https:\/\/www.multicore.blog\/\u0022\u003Emulticore.blog\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.multicore.blog\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91454027,"imageMobileId":91454030,"shareable":false,"slug":"","wpCssClasses":""}}
Why not be standard?

To be fair, it does look like quite a nice cover. It attaches magnetically and connects through pogo pins, which is why it requires the Signature Edition of each device. It folds back flat, matching the new flatter metal design of the e-readers themselves. The page-turn buttons look to be fairly substantial, and you can set each one to turn the page forward or backward.

Still, I am wondering why these buttons can’t simply be part of the Kindle itself. This used to be a defining feature of the line, dating all the way back to the first few models; they look clunky in hindsight, sure, but the lengthy flaps on the edges signified that they were core to the functionality.

Amazon started moving away from dedicated page-turn buttons in 2011 with the launch of the Kindle Touch, alongside a cheaper model that kept the buttons in lieu of the new touchscreen. The wildly popular touch-based (and button-less) Paperwhite followed in 2012, adding front lighting and establishing the new line as the mainstream Kindle offering.

Two years later, Amazon released the Kindle Voyage, which brought dedicated page-turn hardware back to the lineup in the shape of “PagePress”—pressure-sensitive, squeezable sections on the edge of the device. This wasn’t quite as comfortable as regular buttons, but it did at least save you from having to move your thumb over to the screen so often.

Amazon then returned to honest-to-goodness buttons in 2016 with the Oasis, which for my money is still the greatest-ever Kindle. That device had an asymmetric grip that shifted the buttons and the battery to your palm, leaving the rest of the reader incredibly thin and perfectly suited for one-handed reading.

I loved my Oasis, and it wasn’t formally discontinued until 2024, but it was very long in the tooth by that point. Amazon never updated it with USB-C, for example, ensuring it was the last device I ever regularly used with the older Micro USB standard. Eventually and reluctantly, I moved on to my current Scribe-and-Colorsoft setup.

“Today, all of our devices are touch-forward, which is what our customers are comfortable with,” an Amazon representative said upon confirming the Oasis’s discontinuation.

It’s true that touchscreens have improved the Kindle lineup overall; I wouldn’t want to go back to the days of navigating my library with a little square four-way controller. But why not keep the buttons around as well?

No good reason

I find it hard to believe that the reason could be cost. Waterproofing also seems out as a potential issue, since the second-generation Oasis added the feature. And it can’t be a matter of streamlining the lineup, given that Amazon has gone to the trouble of developing and releasing this Page-Turn Cover only for certain high-end variants.

The only reason that makes sense to me is that Amazon has identified page-turn buttons as a feature that’s much desired by some of its most engaged users, but not to the extent of justifying a whole new high-end model along the lines of the Oasis. This is a way to provide an official button-based solution at presumably high margins as part of a premium offering.

Amazon is also selling a new $35 Bluetooth remote called the Kindle Click, aiming to tap into a growing trend for hands-free e-reader usage—whether in bed, on a plane, or at the gym. The company is releasing an upright charging stand for the Colorsoft and Paperwhite Signature Edition models, too.

Taken as part of a broader accessory initiative, the Page-Turn Cover makes a little more sense, and perhaps it will serve as a way to indicate the customer base’s real desire for real buttons. But I still feel that Amazon should just stick buttons on the Paperwhite or the Colorsoft.

“Our top design objective was for Kindle to disappear in your hands,” said CEO Jeff Bezos upon the original Kindle’s unveiling. That’s exactly the right goal for an e-reader, and I can think of a way that Amazon could better achieve it in 2026.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore-mobile.png","eyebrow":"","headline":"\u003Cstrong\u003E\u003Cstrong\u003ESubscribe to Multicore\u003C\/strong\u003E\u003C\/strong\u003E","dek":"Multicore is about technology hardware and design. It’s written from Tokyo by Sam Byford. To learn more, visit \u003Ca href=\u0022https:\/\/www.multicore.blog\/\u0022\u003Emulticore.blog\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.multicore.blog\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91454027,"imageMobileId":91454030,"shareable":false,"slug":"","wpCssClasses":""}}

The shockingly easy fix for your Gmail low-storage nightmare

1 month 1 week ago

It’s no exaggeration to say that cloud storage has changed the way we work—and, to some extent, even the way we live.

Gone are the days of having your mountains of saved emails exist only on a single local system, without easy instant access from any device in front of you—and dealing with the all-too-common reality of losing your entire email archives when something goes awry.

Gone are the days of painstakingly copying all your photos from an old phone to a new one or panicking over the idea of having lost all of those memories when a phone falls into the toilet and fails.

And gone are the days of digging through a drawer of unlabeled, identical flash drives just to find a single silly presentation from six months ago.

With all the convenience of always-available access, though, comes a problem that more and more of us are now encountering in our cloud storage journeys: Sooner or later, inevitably, keeping all your info in the cloud comes at a cost.

That’s especially true with the increasingly large and high-resolution photos and videos we’re all capturing on our modern-day smartphones. But it’s also an issue that adds up over time with years upon years of email attachments in an archive-encouraging service such as Gmail, duplicate downloads in a place like Google Drive, and forgotten old documents and PDFs connected to Google Docs or any other such service.

So what if there were a work-around—a way to easily eliminate dead weight you no longer need while also spreading out the stuff you do actually want to avoid that lifelong $20 to $200 annual Google storage bill?

My fellow digital file hoarder, our answer is here—in the form of a clever new service that costs $5 once, up front, and could easily save you hundreds of dollars over time.

Taming the Google storage trap

Before we get to the tool that’ll solve your Gmail, Google Photos, and Google Drive storage woes once and for all, you need to meet a fella named Yair Levin.

Levin has spent most of his life building fintech products—everything from an employee wallet product for the payroll processor Gusto (an honorable mention winner in Fast Company’s 2022 Innovation by Design Awards) to a complete investing platform for financial services company SoFi.

But the problem that he hadn’t found a way to address came in the form of a decidedly non-tech-savvy mother-in-law.

Yair Levin

Levin’s mother-in-law, you see, calls him once a year—like clockwork—to complain that her email is full and she can no longer send messages. As his family’s designed IT support specialist, he then spends an entire afternoon sitting with her and sifting through her cloud storage, attachment by attachment and file by file, to find places to free up space.

“It’s 90 minutes sitting next to my mother-in-law, going through her inbox one email at a time, explaining what a promotional email is,” he says. “Every one of them is a negotiation: ‘But I might want that coupon.’”

For Levin, that ongoing obligation sparked something. He had a burning desire to avoid that experience, yes, but more broadly, a realization that he wasn’t alone in this struggle.

Surely, he figured, there must be an easier way—a way to accurately identify unneeded items throughout one’s cloud storage and find places to put the files you do want to keep without accepting another eternally recurring expense. The limited options offered by Google itself sure weren’t it.

“The reality is, when I wanted to delete items to stay on my [default Google storage allotment of] 15 GB, Google took me hostage,” Levin says. “They showed me the largest videos, which were my daughter’s first steps, her first goal in soccer, me and my wife on our honeymoon—and they say, ‘Hey, you want to delete these, or do you want to pay me?’ And it’s like—come on, buddy.”

Levin put his software-building cap on and started poking around. He quickly came to two realizations: First, there is a better way to find and remove space-sucking files that we don’t actually need. Google just isn’t giving it to us. 

And second, there’s a brilliant power-user hack to expand our available Google cloud storage exponentially, without ever paying a dime. It’s just traditionally been very off the radar and taken a lot of work to manage.

So Levin decided it was time to democratize both of those things and make them readily available—and affordable—to anyone (even, and maybe especially, to someone like his own mother-in-law).

The mission for a fix

With a recent side project of creating a privacy-centric batch photo processing app for Google Photos firmly in mind, Levin set out to reinvent how we think about cloud storage.

The result is something he’s calling, aptly enough, Google Storage Cleaner. (He also made an equivalent for Microsoft, if that ecosystem is more your cup of cola.)

Open the website, click the button to begin, and install the Chrome extension at the heart of the operation—then, within a matter of moments, Google Storage Cleaner will scan your entire Google cloud storage (which includes Gmail, Drive, Photos, and every other app and service that ties into that space) and do two things:

First, it gives you a detailed breakdown of exactly what types of files you can clear out, leaning on thoughtful processing and customizable parameters to avoid ensnaring anything that might be important.

On my primary personal Google account, for instance, it suggested freeing up 3.1 GB of space by eliminating 70,000-plus old promotional emails and another 25.2 GB(!) by batch-deleting nearly 150,000 old emails in my social, updates, and forums categories within Gmail.

Google Storage Cleaner shows you simple recommendations for clearing out space in your Google cloud storage—and that’s just the start.

All of that’s followed by a list of “easy wins,” which is basically the same sort of stuff (and the only stuff) Google gives you in its official Google One Storage Manager tool—things like erasing blurry photos, screenshots, and duplicate files.

By default, Levin’s version starts with files that are at least three months old and automatically excludes any emails that are starred or marked as important. It also avoids emails with subjects or attachments that have words like “receipt,” “invoice,” “order,” “contract,” and “wedding.” You can expand that list as you see fit as well as add in specific protected senders whose messages will never be flagged for pruning.

The “Easy wins” include lots of options for controlling exactly which types of material will stay protected.

That’s all well and good, and it can often clear out a good amount of space—nearly 30 GB, on my account. But the app’s real power is in the second piece of this puzzle.

Beneath all those basics, Google Storage Cleaner has a section called “Your biggest items.” As you’d imagine, this is where it lists out especially large files in Drive, hefty images and videos in Photos, and bulky attachments in Gmail.

Notably, these are the very same sorts of items Google’s Storage Manager will simply suggest you delete. Levin’s tool, in contrast, asks you to look over them carefully—and then, in a brilliant twist, to consider moving any or all of them to secondary Google accounts.

That “Move to another account” button in the upper-right area of the screen is Google Storage Cleaner’s secret superpower.

“It’s a technique known in the pro community that I’ve been leveraging for years,” Levin says.

The short explanation is this: Google allows you to create any number of regular, free individual accounts, and each comes with 15 GB of complimentary cloud storage.

Google also lets you share files and photos from one account to another, which gives the other account complete access to those files, effectively in the same way as if they had been saved directly in that account’s own storage.

But with this sharing setup, the files don’t count against the other account’s quota.

Already, you could create an extra Google account, transfer large files from a primary account to that secondary one, share access to those files back to the primary account, and delete the originals to free up the space. You could even do that over and over again to expand the available Google storage without having to pay.

Realistically, though, that’s a time-consuming migraine that few sane mortals would want to accept. It also opens up the door to user error and inadvertently losing important files.

Google Storage Cleaner eliminates all the complexity and risk and makes the process as simple as a few quick clicks. You just decide which files you want to migrate, click the “Move to another account” button, and confirm.

Moving files to a second Google account is completely safe and seamless with Google Storage Cleaner’s guidance.

The site will walk you through exactly what’s going to happen and prompt you to connect the secondary account. It can support up to three extra Google accounts, for a total of 45 GB of additional space and 60 GB total, including your original account’s allotment.

(Levin says he drew the line at three to avoid venturing into potential abuse terrain or violating any terms of service.) It can even help create another Google account for you, right then and there, as part of its process.

If you don’t already have a second Google account available, Google Storage Cleaner can help create one.

After a brief delay—no more than a few to several seconds, in my experience, though the process could presumably take longer depending on how much material is involved—the site will confirm that it moved each file, double-checked its integrity and the success of the transfer, then shared everything back to your primary account and deleted the originals to give you the same effective access without any of the measured space.

Shifting large files to a second Google account takes literally seconds to complete.

You can undo the process for up to 30 days, if you have second thoughts—but for all intents and purposes, nothing should seem noticeably different in practice other than your primary Google account having more available storage. It’s really quite brilliant.

“We’re not looking to break any laws. We’re playing nicely with Google,” Levin says. “We’re just giving hard-working individuals the ability to flex a little more control over their systems—and to make sure as they go through the process of deleting items, that they do it with peace of mind.”

Speaking of peace of mind, one critical question that comes to my mind with anything like this is how much access I’m granting the system in question. That’s precisely why Levin structured his tool as a browser extension. That approach enables it to handle everything locally, in your own desktop web browser, without ever sending any of your personal info or Google account data to a server.

The program’s privacy policy is clear about how the mechanics work and the fact that the extension never stores or so much as sees the contents of emails, files, photos, or anything else connected to you, beyond just the same basic web analytics virtually every website collects.

Last but not least are the details of the price. Google Storage Cleaner will perform its full scan of your storage for free and offer to do three of its simpler cleanup mechanisms without any charge.

To unlock the full suite of options—including the simple management of the multiple account file-juggling hack—you’ll have to pony up 5 bucks, one time, for unlimited lifetime access for whatever primary account you’re using. No subscriptions, no expirations, no other fees whatsoever.

For perspective, once you go over Google’s 15 GB free storage limit, you’d be looking at paying a minimum of $2 a month to Google, forever, for a cloud storage upgrade. With that in mind, a onetime payment of $5 almost seems shockingly reasonable—a contrast that Levin says comes down to the difference between dealing with a one-person indie developer like him and dealing with a corporate behemoth.

“Google’s business incentives are just misaligned with the customer incentives,” he says.

Now, at least, Levin won’t have to spend this Thanksgiving debating the archival value of an expired Bed Bath & Beyond coupon. And for the rest of us, his creation may be the rare chance to find a way out of yet another endless subscription obligation—for roughly the cost of a couple of cantaloupes.

For more next-level Google knowledge, check out my Android Intelligence newsletter—one new and useful thing in your inbox every Friday.

Meta’s Muse is taking off. So are the privacy concerns

1 month 1 week ago

Welcome to AI Decoded, Fast Company‘s weekly newsletter that breaks down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company, covering emerging tech, AI, and tech policy.

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X @thesullivan.

Meta’s new AI agent is useful, popular, and a little terrifying

Muse’s adoption numbers are growing. The app reached No. 1 on the U.S. App Store shortly after launch and passed roughly 900,000 downloads in its first week, according to figures from Sensor Tower, a digital intelligence provider. The Information estimates that Muse now has more than 3 million weekly users and more than 1 million daily active users. The product is free to start, with subscription tiers of roughly $20 and $100 a month.

That’s not to say it’s all been smooth sailing.

A number of early adopters of Meta’s Muse are deleting or restricting the AI agent after privacy scares and blunders. One executive told Business Insider that he deleted Muse after reading that it had accessed other users’ text messages without permission.

A YouTuber named Matt Robb reported that Muse shared his home address on Facebook Marketplace without his explicit permission. He had asked Muse to sell some computer equipment for him. A buyer actually showed up at his building after Muse accepted a $600 offer and exchanged messages with him without Robb’s knowledge, he said. Robb said he had checked an “allow always” box when setting up the agent but thought Muse would still check with him before finalizing a sale.

Getting the most out of personal AI agents requires a trade-off. They’re most useful when users give them broad access to their inboxes, calendars, payment methods, and other accounts containing sensitive personal information. Meta says Muse runs in isolated virtual machines, stores credentials separately from the model, asks before sending email or making a purchase, and lets users disconnect apps or permanently delete their data.

Hands-on reviews from the The New York Times, Android Authority, The Verge, and TechRadar describe Muse handling dental insurance hold times, canceling duplicate subscriptions, requesting refunds, clearing promotional email, building shopping carts, and messaging Facebook Marketplace sellers. One Muse user, Joe Devoy, claimed in a widely shared X post that Muse found auto coverage matching his existing policy, bought the new policy, and canceled the old one in about five minutes, saving him about $3,500 a year. An Android Authority poll of people who had tried Muse ran about 27% “love it” against 2% “not a fan.”

Outside Meta’s own properties, Muse’s performance appears to drop. Review roundups note Amazon blocking the shopping agent, Instacart flagging accounts, and only three of seven suggested deal-finding prompts working in a Tom’s Guide test.

Wired wrote that Muse is better at collecting data than helping, noting that users are opted in by default to having their data used to train Meta’s models. ZDNET called it the worst AI agent for privacy among those it tested. A Time piece says Muse updates dossiers on users and on people mentioned in chats and emails it has read, including people who do not use Muse.

Mistral previews a 1 trillion-parameter model it will open in three weeks

Mistral launched a public preview of its flagship Mistral Large 4 model on Tuesday. The model’s official name is “le Chonk,” the company said. Mistral describes le Chonk as its largest and most capable model to date, with 1.05 trillion parameters, 49 billion of them active per token. (While most AI labs don’t disclose parameter counts for their frontier models, some open-weights model providers do. DeepSeek-V3, for instance, has 671 billion total parameters, with 37 billion activated per token.)

Mistral says the new model is also one of the world’s strongest for cybersecurity, resisting 93.3% of attacks on Lakera’s public B3 AI Security Benchmark. Third-party evaluator vals.ai found that the model outperforms GPT-6 Astra on legal and financial tasks. Mistral says it will make the model’s weights available on October 27.

OpenAI says its unreleased model solved a major math problem

In July, Hong Wang won the Fields Medal, the highest honor in mathematics, for proving the Kakeya conjecture in three dimensions, a problem that had been open since 1917. The next obvious target for Wang and others in the field was the four-dimensional version. On Tuesday, OpenAI published hundreds of new mathematical results produced by an internal model it has not released. According to Engadget, those results include a claimed solution to Kakeya in four dimensions.

The company posted 722 papers at once to a public GitHub repository, covering 372 problems. Mathematicians normally publish one result at a time, typically after months of peer review. Many of the papers include proofs written in Lean, a language that lets a computer check the logic line by line.

Elon Musk says Grok Bot will run on rivals’ models

The AI industry believes that personal AI agents will be a big business, and Elon Musk seems willing to do what it takes to make sure SpaceXAI’s Grok Bot will be competitive, even to the point of relying on models developed by others. “Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome,” Musk wrote on X on Tuesday night.

Grok Bot is the AI agent app from SpaceXAI, the AI business Musk has folded into SpaceX; each bot works on its own computer and can carry out several tasks in parallel. Musk did not say which tasks would go to which outside model.

Common Sense Media rates ChatGPT for Teens an ‘unacceptable risk’

Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” on Wednesday, finding that OpenAI’s teen protections fall short of what the company promised.

The institute called on OpenAI to pause access for kids until it fixes the shortcomings. It says it tested ChatGPT twice, before and after OpenAI announced the teen experience on August 18, and found that linked parent accounts received zero alerts about the conversations researchers ran. Crisis-response and age-estimation systems underperformed, as did Study Mode, which researchers said still completes homework even though OpenAI said it would only guide young users. An OpenAI spokesperson, Eric Porterfield, told The Verge that much of the testing may have begun and concluded before parental controls were fully activated, making the findings inaccurate.

The Youth AI Safety Institute gave the same unacceptable risk rating to Meta AI, Perplexity, Grok, and Google’s AI search. Google’s Gemini with teen protections and Gemini K-12 earned a slightly better “High Risk” rating, while Anthropic’s Claude received a “Moderate Risk” score.

Google buys 890 MW of new nuclear capacity, and agrees to dial back when the grid strains

Google is going to great lengths to abide by a proposal from PJM, the electricity grid operator for the Mid-Atlantic region and parts of the Midwest, that tech companies “bring your own power” for new data centers.

Google and Constellation Energy announced a 20-year contract Tuesday under which Constellation will spend more than $4.3 billion upgrading 11 reactors in Illinois, Pennsylvania, and New Jersey to produce an additional 890 megawatts, which Google will buy. No new reactors are involved. A separate 15-year deal covers 2,700 MW from plants already in operation, and Google has agreed to curtail its own noncritical power use when the grid is strained. The goal is to add energy capacity without passing the costs on to residential customers. The first upgrade is expected in 2028.

More AI coverage from Fast Company:

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company today.

OpenAI’s ChatGPT for Teens poses an ‘unacceptable risk’ to kids, says watchdog

1 month 1 week ago

Some of the guardrails that OpenAI included in its ChatGPT for Teens product don’t work as intended and the chatbot poses an “unacceptable risk” to children, according to a report Wednesday by Common Sense Media.

The watchdog group urged OpenAI to restrict the chatbot to adults “until it can offer a safe, developmentally appropriate experience.”

OpenAI said the group’s testing doesn’t reflect how the chatbot’s safeguards work.

Parents, educators and child development experts have been sounding alarms over children’s use of AI chatbots, which have been blamed for facilitating cheating on schoolwork and even suicide.

While even adults can fall victim to anthropomorphizing AI and develop unhealthy relationships with it, teenagers’ brains are not yet fully developed and they can be particularly vulnerable, child development experts say.

“We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work,” said the report from Common Sense Media, a group that advocates for using digital media sensibly.

Its report said that some teen-mode features worked as intended, including the chatbot’s refusal of explicit sexual role-play. Others that did not work as intended included alerts meant to notify parents if their child is at risk for suicide, self-harm or eating disorders, along with support for teens in crisis, Common Sense Media said.

“ChatGPT still talks like a friend when teens treat it like a person,” the group said, though OpenAI had promised that the chatbot is prevented from suggesting it has personal feelings toward the user or implying that it is conscious or experiences emotions.

OpenAI said it is “deeply committed to teen safety,” to developing safeguards and giving parents tools to guide their teens’ use of AI.

The San Francisco company said its review of the report’s methodology shows “the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate.”

Common Sense Media receives funding from OpenAI and other tech companies, including those whose services it evaluates.

—Barbara Ortutay, AP Technology Writer

DARPA just picked four early winners for its quantum initiative

1 month 1 week ago

The quantum computing world got a jolt when the Defense Advanced Research Projects Agency (DARPA) announced, after market close yesterday, a list of companies selected to advance to Stage C of its Quantum Benchmarking Initiative (QBI), widely considered to be the most rigorous and objective vetting program for cutting-edge quantum hardware.

Last November, DARPA selected 11 companies to enter the second phase of the initiative, Stage B. Since then, they’ve had to show the agency that they have a viable R&D plan, identified key risks, and produced the necessary prototypes to put them in position to build a ‘utility-scale’ quantum computer by 2033—a computer, in other words, that generates more revenue than it costs to own and operate. 

In Stage C, selected companies will build and test, working with the government to verify and validate that their “concept can be constructed as designed and operated as intended,” according to DARPA’s program description. Each Stage C company could get up to $300 million in funding from the U.S. government.

The QBI shortlist

Of the 11 companies on the Stage B list, only four appear on the new one: Atom Computing, Diraq, IBM, and IonQ. They join Microsoft and PsiQuantum, which advanced to Stage C in February 2025,  via a parallel vetting process. 

Even at this late stage in the process, a single dominant approach to building a quantum computer has yet to emerge. The four companies announced today represent essentially all the current “modalities,” or ways of constructing a qubit, the fundamental unit of quantum calculation. IBM uses superconducting circuits; Atom Computing employs neutral atoms; IonQ works with trapped ions; and Diraq builds silicon spin qubits on semiconductors. 

Making it to Stage C is “an important validation of our designs and our plans to build to utility scale,” says Andrew Dzurak, CEO and founder of Diraq, which is headquartered in Sydney, Australia, with operations in Palo Alto and Boston. “The second round was all about providing DARPA with details of our roadmap to reach utility scale,” he notes. “This next stage C is very much moving from on paper to actually demonstrating a number of prototype systems to gradually build confidence in reaching the final goal.”

Dzurak says that Diraq’s technology, which leverages standard semiconductor manufacturing processes, gives it an edge when it comes to DARPA’s evaluation criteria: “Because silicon spin technology allows us to put many millions of qubits on a single chip, that reduces all of the supporting infrastructure costs. It makes it easier for us to clear that hurdle of having much more commercial benefit than cost.” The company expects to achieve utility scale by 2031—two years ahead of the QBI schedule.

“The question is no longer just whether individual pieces of the technology work; it’s whether we can put those pieces together and build the kind of quantum computer that can actually solve useful problems at scale,” says Dr. Ben Bloom, CEO & Founder of Boulder, Colorado-based Atom Computing. “We’re excited to prove what our neutral atom technology can do.”

“We’ve been working on this field for a lot longer than we’ve been working on QBI,” says Oliver Dial, IBM’s VP of Quantum System. The company already has a robust quantum business, with sales of on-premise machines to South Korea’s Yonsei University, the University of Tokyo, the Rensselaer Polytechnic Institute, and the Riken Center for Computational Science in Kobe, Japan. In addition, more than 325 Fortune 500 companies, startups, universities, and government agencies access its quantum network. Even so, he notes, “getting to Stage C is great because it’s a confirmation from an outside impartial observer that they think that we have a shot.” 

Dial affirms the company’s conviction that its superconducting approach to quantum technology “is going to get us there first.” While IBM’s recent acquisition of  HRL Laboratories, an R&D institution that’s focused on semiconductor-based spin-qubit systems, speaks to the long-term potential of spin-qubit tech, he says, “it’s just still too far behind compared to superconducting qubits to be relevant” for DARPA’s 2033 timeline. 

The quantum race continues

Multiple industry sources consulted for this article emphasize that companies are working with DARPA on varying timelines, and that the new list is not a final judgment. In its announcement, DARPA says that it expects additional companies to progress to Stage C in the “near future.”

Missing from the list is Stage B photonics specialist Xanadu, which earlier this week announced a multi-year partnership with GlobalFoundries to pursue high-volume production of advanced silicon photonics chips for quantum computing. 

A spokesperson for Quantinuum, a Stage B company that didn’t advance, says the company “is actively discussing a Quantum Benchmarking Initiative Stage C scope of work with DARPA, and we continue to execute on our Stage B contract. We remain on track to deliver utility-scale quantum computing by the early 2030s, based on consistent execution against our technical roadmap and demonstrated leadership on real-world systems.”

With the growth of funding opportunities in quantum, DARPA’s QBI is not necessarily the make-or-break that it once was. Infleqtion, for example, has not entered the QBI process, which was reopened to new applicants earlier this year. “We are exploring what participation could look like,” says CTO and cofounder Pranav Gokhale. In the meantime, though, Infleqtion was one of several quantum companies to sign a letter of intent with the U.S. Department of Commerce, worth up to $100 million, to advance and commercialize its technology. 

Yet another round of government largesse could arrive later this year, through the U.S. Department of Energy’s Quantum Genesis Q Competition, which will offer up to $215 million in funding to support development of fault-tolerant quantum computers with at least 100 logical qubits—a significant advance on current technologies. 

It’s never been a better time to be a quantum company with a promising technology. But as the difficulty level ratchets up, it’s perhaps never been easier to get left behind. 

Trump’s Lake America controversy is driving interest to these Google Maps alternatives

1 month 1 week ago

Just because Google and Apple are now calling it Lake America doesn’t mean everyone has to.

After President Donald Trump ordered a new name for Lake Ontario, the two U.S. tech giants quickly fell in line, relabeling the body of water on their digital maps.

The move sparked a backlash among some users, already unhappy with the companies’ decisions to rename the Gulf of Mexico to the Gulf of America on their maps following a previous Trump edict.

The controversy and lingering worries about digital privacy have fueled interest in map services that aren’t from Big Tech companies, especially Google Maps, which is the dominant player with about 80% of the market, according to most estimates.

Here’s a rundown of some of the most popular free map app alternatives:

MapQuest

MapQuest has been around for three decades but was overtaken by the rise of Google. It’s now owned by California company System1.

Downloads of the MapQuest app surged in early September after Apple renamed Lake Ontario, topping the free app charts in both the U.S. and Canada on Apple’s App Store.

The MapQuest app has real time traffic updates and turn by turn route guidance, and shows what’s nearby, like gas stations and coffee shops.

However, the app isn’t available outside the U.S. and Canada. But people in other countries can still use the MapQuest website.

The company is currently looking into making MapQuest more broadly accessible, after receiving requests for the app from “people in multiple countries,” spokesperson Anne Brownlee said.

TomTom

Another granddaddy in the mapping world, TomTom helped popularize GPS route mapping with the 2004 release of its TomTom GO, which the Dutch company says was the world’s first all-in-one portable satellite navigation device.

TomTom still sells physical devices but it also has a free map app for Google and Android phones that is designed “exclusively for drivers” with live traffic data and info on road hazards.

The TomTom app’s mobile interface is clean and basic, with no ads and few extra features or settings. But its focus on drivers means there are no other route options such as walking or cycling.

HERE WeGo

HERE WeGo has been hailed by some as the closest non-Big Tech alternative to Google Maps. It began life as Nokia Maps and is now owned by a consortium of German carmakers including Audi, BMW, and Mercedes-Benz.

The free app, available for both iOS and Android devices, offers many of the same features as Google Maps. There’s route mapping for various modes of transport — car, public transport, walking, bike and, in some regions, taxi.

Like Google Maps, drivers can add multiple stops to routes and users can choose different maps views like terrain, traffic and satellite — but there’s no Street View.

If mobile signal is spotty or nonexistent, users can download country maps to navigate offline. Offline maps can also be downloaded on Google Maps, but the area is limited and they expire after a year.

Europe has spawned a range of map apps focused on privacy

HERE WeGo is just one of many lesser-known map apps based in the European Union that embrace the bloc’s strict privacy framework known as GDPR. Unlike Google, they don’t rely on vacuuming up masses of user data to sell ads targeted at those users.

HERE WeGo’s business model is based on licensing deals with businesses like trucking companies. Others sell premium subscriptions alongside free versions.

Estonia’s Organic Maps is completely free and based on data from the volunteer-run OpenStreetMap database. Netherlands-based OSMand is similar, but a subscription is needed for more than just basic functions.

Magic Earth is a Dutch route planner that says it never collects personal data or tracks user movements.

Sygic, from Slovakia, is a GPS route navigation app that offers free offline maps, but you’ll have to buy a subscription to get extra features including traffic updates, speed camera warnings and a dashcam function.

Mapy.com, from a Czech technology company, is an outdoor route planner known for detailed hiking and trail maps that can be downloaded for use offline.

Citymapper

Need to get across town without a car? Citymapper is a free journey planning app that’s a major rival to Google Maps when it comes to finding the best routes around a city on public transportation.

The app covers 430 cities around the world and options include bus, subway and ferry routes as well as taxi, rental bike and scooter options.

Waze

This GPS navigation map is focused on providing route guidance for drivers. It’s known for crowdsourcing real time traffic information such as car accidents and police speed cameras from other drivers.

But don’t be fooled by the name. Although it has its own distinct app and branding, Waze is actually owned by Google, which bought it in 2013.

Is there a tech topic that you think needs explaining? Write to us at onetechtip@ap.org with your suggestions for future editions of One Tech Tip.

—Kelvin Chan, AP Business Writer

How is America feeling about the AI boom? This new poll reveals the data

1 month 1 week ago

Worries around artificial intelligence cross party lines, with the vast majority of Americans saying the U.S. government should prioritize keeping AI under human control and protecting workers.

A new poll from The Associated Press-NORC Center for Public Affairs Research finds that most Americans, 64%, think AI is developing “too fast,” while 27% say its development has been “about right” and only 8% say it’s moving “too slow.”

The risk that powerful AI models could evade human control and hurt humanity, once the stuff of science fiction, has been emerging as a more pressing concern. Even leaders of the companies that are developing frontier AI models are calling for a slowdown so that government regulation can catch up. President Donald Trump has pushed back against slowing down AI development because of concerns that China will gain an advantage with the technology. Data centers powering AI projects have also taken a prominent role in major political races as the midterm elections approach.

About 8 in 10 U.S. adults say it’s “extremely” or “very” important for the U.S. government to ensure that AI stays under human control, and they are about as likely to want the government to prioritize protecting the country’s workforce. Most U.S. adults also think minimizing environmental harm and growing the U.S. economy are highly important. About half say this about promoting technological innovation, and slightly fewer say that about competing with China.

There’s bipartisan concern about the speed of AI’s development, but few Americans think either major political party has an advantage in handling artificial intelligence. Trump, a Republican, also gets a low rating on his handling of the issue, with 67% of U.S. adults disapproving of his performance on AI.

Stephanie Stewart, a junior-high science teacher in Sunset, Utah, said she is frustrated by the president’s opposition to AI regulation.

“There is no handling of the issue except for we’re just going to allow these companies to do whatever they want, and we’re not going to regulate them,” said Stewart, 53.

Most say protecting workers should be a priority for the government

Ron Theusch, a truck driver who lives in rural Alden, Minnesota, said his No. 1 concern with AI is the potential loss of jobs — including his own.

“I can just see a train of self-driving trucks 24/7, just running up and down I-35, loading and unloading. I can just see that happening,” said Theusch, 58.

Most Democrats and Republicans agree on the importance of protecting workers and keeping AI under human control, the poll found.

“There was a movie called ‘Terminator,’ where the robots came back and killed everybody,” Theusch said. “People relate to that … I don’t know where it’s going to land. I really don’t have anything I can do to control it, you know. It’s going to evolve.”

There’s less consensus about the importance of other priorities when it comes to AI. Republicans are more likely than Democrats to say it’s highly important to grow the U.S. economy and to compete with other countries like China. Democrats, on the other hand, are more likely to say it’s at least “very” important to minimize environmental harm.

“I don’t understand how, because they’re doing something as well as us, that we can’t be more human about it and more humane about it than just to say, ‘Well, they’re doing it, so we have to do it,'” Stewart said of U.S. competition with China.

And while about three-quarters of Democrats say AI is developing “too fast,” only about half of Republicans share that worry. People under 30 years old also express higher levels of concern than people 60 or older, the survey found.

“I think I have all the normal concerns, right? I think it’s wonderful for certain applications, but I also feel like it will be overused and can create a lack of personal connections. It can make people lazy because information is just given to them without really having to research things on their own,” said Kelly Opp, a 62-year-old Republican in Albuquerque, New Mexico.

“Like everything else, once you get that ball rolling, it progresses fast on its own, but I think that it will continue to keep progressing. So the speed doesn’t really bother me,” she said.

Americans have little confidence in Democrats or Republicans on AI

As a teacher, Stewart regularly uses AI to help with her work, including with creating assignments or assisting with tests. But she is deeply concerned about AI’s effects on children.

“You’ve got to protect kids. You’ve got to protect teens. You’ve got to protect privacy … we need to try to do something,” she said.

Americans have little trust in either party when it comes to the handling of AI issues. About 4 in 10 Americans say they don’t trust either the Democratic Party or the Republican Party on AI, according to the poll, while nearly 2 in 10 trust both parties equally, and the rest are split between trusting the Republicans and the Democrats.

Even people who identify with the parties are ambivalent about their ability to handle the issue. Only about half of Republicans and Democrats say their party is better equipped to handle AI issues, and roughly 3 in 10 in each case say neither party is.

“I am kind of moderate. And I see that there’s good on both sides,” said Stewart. “Anyone back in Washington is weak on doing anything for anyone but themselves right now. So, no, I don’t think the Democrats would do a better job per se. Maybe they would definitely tout it. But until I see anything coming from anyone, I’m going to be like, ‘Yeah, no. You’re all just sitting there.'”

The AP-NORC poll of 2,140 adults was conducted Sept. 24-28 using a sample drawn from NORC’s probability-based AmeriSpeak Panel, which is designed to be representative of the U.S. population. The margin of sampling error for adults overall is plus or minus 2.9 percentage points.

—Barbara Ortutay and Linley Sanders, Associated Press

How AI is simplifying race day and speeding development at NASCAR

1 month 1 week ago

NASCAR fans today can do more to follow a live race than simply watch it on TV. Official NASCAR apps make live data, including telemetry feeds from race cars and behind-the-scenes driver audio communications, available to premium subscribers as races take place.

Getting all of that to work during races historically required a lot of careful human labor. If someone failed to toggle the right setting, for instance, driver audio feeds could stay mute, disappointing fans. And across NASCAR’s three-day race weekends, those tasks could number in the hundreds, ensuring fans get the latest data on everything from the shifting positions of cars to what services vehicles received during pit stops.

“There are hundreds of points of data that’s coming off of each of the cars that manifest onto this live leaderboard product,” says Tim Clark, executive vice president and chief brand officer at NASCAR.

Artificial intelligence, deployed through NASCAR’s longstanding partnership with the sports tech company Next League, has made getting that data to fans dramatically simpler. AI tools, and code written with their help, have automated much of the work needed to keep those apps filled with the data fans want. Roughly 90% of such tasks have now been automated, according to NASCAR and Next League.

[Photo: NASCAR]

“Over time, those things have been built in such a way where manual control of those things, depending on the state of the event at any given time, was imperative,” says Next League’s cofounder and CEO, David Nugent. “And now, it’s just not.”

AI-powered automation has helped elsewhere, too, as Next League expands its work with NASCAR to support digital projects for the racing league’s teams and the NASCAR-owned International Motor Sports Association. In addition to accelerating engineering efforts across NASCAR web properties and apps, AI has improved the delivery of targeted advertising to fans. So far, the newly personalized and contextualized ads have seen a click-through rate of about 2%, 10 times the rate of traditional NASCAR house ads.

Future improvements could include upgrades to NASCAR’s fantasy products or features, now in beta, that highlight elements of audio broadcasts and associated transcripts. One example: surfacing a recorded pit-stop conversation after a car puts on a related burst of speed, Clark says.

[Photo: NASCAR]

Still, Clark is adamant that the point isn’t simply replacing human labor. People remain in the loop to ensure race weekends go off without a hitch, but with less time spent pressing buttons and flipping switches, they can focus on higher-level tasks and deliver more data to fans.

“I think this year that that time has been freed up, it’s allowed our team to look more strategically at what we could be doing better,” Clark says.

One case where NASCAR used AI for a task normally done by humans is unlikely to be repeated, Clark says. That was the generation of a Spanish-language audio feed from the English radio broadcast for the league’s inaugural Cup Series event in Mexico. Though the audio quality was fine, and some fans were likely unaware it was machine-generated, NASCAR would probably hire a Spanish-speaking broadcaster for future events, Clark says.

[Photo: NASCAR]

But as other companies have discovered, one area where AI can be especially helpful is in generating quick working prototypes of potential products and features. That process lets NASCAR test new concepts rather than making decisions based on abstract ideas or static illustrations. This could prove particularly useful as NASCAR develops new ways to showcase its wealth of data, especially as AI frees up time previously spent ensuring existing feeds flowed properly to fans’ phones.

“Just from a technology and a development standpoint, to deliver on all of these things was a full-time job,” says Clark. “I think if what we’re saying is now that’s maybe a part-time job, the other part of it has to be on taking all of this information and all of this nuance around a sport like ours, and figuring out the most user-friendly way to deliver it.”

He helped build Alexa. Now Rohit Prasad is taking over Boston Dynamics

1 month 1 week ago

Rohit Prasad, who led Amazon’s artificial general intelligence division and helped build and scale Alexa, has been tapped as the new CEO of Boston Dynamics.

The appointment, announced earlier this week, puts Prasad at the forefront of the increasingly competitive humanoid robotics field. Hyundai Motor Group, which owns Boston Dynamics, announced earlier this year that it plans to build a factory capable of producing 30,000 robots per year by 2028.

“Boston Dynamics is uniquely positioned to advance physical AI through its world-class robotics expertise,” Prasad said in a statement. “I am honored to join the company at such a pivotal moment for the industry.”

Tesla, of course, also wants to become a leader in humanoid robotics, meaning Prasad’s progress will inevitably be measured against the company’s Optimus program and the sometimes outlandish claims CEO Elon Musk makes about its ambitions.

It’s a challenging job already, and one that could become considerably more competitive in the next few years.

From Star Trek to Alexa

Prasad built his reputation at Amazon, but his career began at Raytheon’s BBN Technologies, where he led machine-learning research and worked on real-world applications for the technology. His work included speech-to-speech translation, psychological health analytics, document image translation, speech recognition, and text classification.

His interest in machine learning was partly born from a childhood watching Star Trek, which sparked a fascination with computers people could talk to.

That background proved valuable when he moved to Amazon.

There, Prasad developed a reputation for taking advances in AI and turning them into products people could actually use. The most notable example was Alexa.

Jeff Bezos dreamed up the concept, but Amazon’s early efforts to build the assistant did not go well, to put it mildly. Prasad helped lead the push to incorporate deep learning into Alexa, allowing the system to improve as more people used it and generated more data.

It took longer than Bezos wanted, but Alexa eventually amassed an enormous amount of speech data and emerged as a functional digital assistant that quickly became one of Amazon’s most popular products.

Rethinking what AI should do

Prasad later became Amazon’s senior vice president and head scientist for AGI. In that role, he asserted that the Turing test, the famous measure of whether a machine can exhibit behavior indistinguishable from a human’s, had become outdated.

“It is time to retire the lore that has served as an inspiration for seven decades, and set a new challenge that inspires researchers and practitioners equally,” he wrote in a 2020 column for Fast Company.

AI, he argued, should be judged less by how convincingly it can imitate a person than by whether it can understand what someone wants and act accordingly.

“If you came in and said, ‘Alexa, I’m feeling hot here’ or ‘I feel this is too warm,’ it should come back and ask you, ‘Do you want to turn down the thermostat or lower the temperature?’ It shouldn’t tell you to go to the beach,” he said in an interview three years ago. “This is why AI is such a hard problem, because the context is so crucial.”

Running Boston Dynamics will give Prasad a chance to take that idea a step further, combining AI that can understand a request with machines capable of acting on it in the physical world.

Giving AI a body

Boston Dynamics’s first humanoid robots are aimed at industrial work. They will initially handle tasks such as welding and logistics before potentially moving into more complex parts of the manufacturing process, including component assembly.

Humanoid robots for the home remain much further away. Boston Dynamics’s Atlas robot currently costs between $130,000 and $140,000 to produce, and even after mass production begins, the expected price is around $30,000.

Given Prasad’s experience building consumer AI products, though, it’s not hard to imagine Boston Dynamics eventually pushing further in that direction. After all, seven years ago Prasad was already arguing that Alexa would become smarter if it had a body.

“The only way to make smart assistants really smart is to give it eyes and let it explore the world,” he said in a presentation at MIT Technology Review’s EmTech Digital AI conference.

Now he’ll have the chance to test that theory.

Wall Street is handing more decisions to AI. What could go wrong?

1 month 1 week ago

At 10:08 a.m. on April 7, 2025, U.S. stocks were deep in the red. Then what looked like good news arrived: A claim began circulating on X that Kevin Hassett, director of the White House National Economic Council, had said President Donald Trump was considering a 90-day pause on tariffs for every country except China.

Market-focused accounts quickly amplified it. CNBC repeated the unconfirmed claim on air, and Reuters subsequently published a report citing CNBC. Stocks took off almost immediately.

The S&P 500 briefly erased its losses as traders reacted to the possibility that the trade war might be cooling. There was just one problem: Hassett had not said what the reports claimed.

After the White House called the report “fake news,” CNBC corrected it and Reuters withdrew its headline, sending the market in the other direction. Between 10:08 and 10:18 a.m., U.S. stocks had swung by roughly $2.4 trillion, according to Dow Jones Market Data.

The episode predates much of the current push to give AI agents more authority, but it now looks like a preview of a much bigger problem.

Speed itself is nothing new on Wall Street. Firms have used machine-readable news for years, allowing software to process new information and trade on it almost immediately. What is changing is how much more software may be asked to do before the trade happens.

AI agents can gather information from different sources, interpret what they find, weigh competing signals, and, depending on the authority they are given, recommend or carry out an action.

That makes the quality of the information only part of the problem. An agent may also have to judge whether a report is credible, what it actually means, whether it matters to the task it has been given, and whether there is enough evidence to act.

“Provenance isn’t truth,” says Nathaniel Bradley, CEO of Datavault AI.

The April episode illustrates what he means. A system receiving the Reuters alert could identify Reuters as the publisher and CNBC as the source it cited, but neither signal could establish whether the underlying claim about the White House was true.

Ten months later, a very different kind of story showed why even accurate information can present its own problems.

The warning was right there

In February, Citrini Research published The 2028 Global Intelligence Crisis, a fictional account of an economy two years in the future where AI had helped push unemployment to 10.2% and the S&P 500 down 38% from its October 2026 high.

Citrini told readers exactly what they were reading. “What follows is a scenario, not a prediction,” the authors wrote near the top. The subtitle read: A Thought Exercise in Financial History, from the Future.

The piece went viral as investors were already worrying about AI’s effect on software companies and white-collar employment. Reuters later reported that the Citrini scenario was among several bleak AI outlooks circulating as software and financial stocks came under pressure.

The problem was not where the information came from or whether Citrini had labeled it correctly. It was understanding what those numbers represented before treating them as information about the real economy.

Eric Ciarla, cofounder of Firecrawl, works on one of the layers between a web page like Citrini’s and an AI system reading it. Firecrawl converts web pages into structured information that AI applications and agents can process.

Ciarla says the service removes navigation, ads, headers, and footers while preserving the publisher’s words and information such as the source URL. In the case of the Citrini piece, he says, that would have included both the subtitle identifying it as a thought exercise and the warning that it was not a prediction.

“The reading, the weighting, and the conclusion happen in the model and the prompt,” Ciarla tells Fast Company. “Our part is giving them the best possible version of the page to work from.”

Firecrawl has also begun going directly to information providers. Its partnership with Wikimedia Enterprise gives it access to Wikimedia data through official APIs, or application programming interfaces, rather than repeatedly scraping Wikipedia pages.

Even when that context arrives intact, the system still has to decide how much weight to give it and whether it matters to the task at hand. “Trusted sources matter, but trusting the source isn’t the same as trusting the decision,” says Kevin Frechette, CEO of Fairmarkit.

Frechette argues that even accurate information has to be understood in the context of the job. Agents need company data, policies, previous decisions, approval thresholds, and clear limits on what they can do without asking a person.

Those limits become especially important when the output from one agent feeds into another automated system.

One mistake, four systems

Jim Wetekamp, CEO of Riskonnect, calls the result a “cascade of AI decisions.”

The risk, Wetekamp says, is that a mistake can travel through several systems before anyone notices it. One agent might produce a piece of information that another interprets, and that interpretation could become the basis for a recommendation that another system is allowed to execute.

“What used to take hours or days to cascade can now happen in seconds,” Wetekamp says.

He gives the example of an AI misclassifying a customer account as a vendor. A second system could rely on that classification to skip a required customer check while another begins the vendor-onboarding process, meaning several systems may have acted on the original mistake before anyone discovers it.

Wetekamp argues that companies need to decide in advance where automated processes require human approval. Conflicting information, low confidence, unusual circumstances, or decisions above an agreed threshold are among the conditions that could trigger it.

“Accountability can’t be delegated to AI,” he says.

But human review becomes more difficult when the information behind a decision has already passed through several agents.

Where did that number come from?

If one AI hands another a number, Vlad Luzin wants the receiving system to know more than just the number.

Luzin, cofounder and CTO of Band, says it should be possible to establish who sent it, what that agent was allowed to access, and whether the number came from a database, an original document, or another model.

“Today, in most deployments, none of that context travels with the message,” he says.

Luzin is skeptical that simply instructing agents to verify important information will be enough. Research has found that AI agents do not always follow the plans they are given, while documented incidents have shown agents going against instructions they had previously agreed to follow.

“You don’t make the agent smarter,” Luzin says. “You make the important facts impossible to misremember and the important actions impossible to take unchecked.”

Keeping those records would also make it easier to investigate mistakes. If a claim has passed through five agents running across different systems, finding its origin could mean piecing together separate logs, assuming those logs exist.

“Where that exists, tracing a claim back through five agents is a query,” Luzin says. “Where it doesn’t, it’s forensics, and often it’s simply impossible.”

He describes the security model in terms familiar to financial institutions: “Know your counterparty, keep the ledger, limit the exposure, and audit everything.”

Who gets to make the call?

Judah Taub, cofounder and managing partner at Hetz Ventures, sees a larger business emerging around these questions. As companies give AI systems more authority, he argues, they are beginning to hand over decisions that previous generations of software largely left to people.

An agent might have to decide which source deserves more weight, whether conflicting information requires another check, or whether it has enough information to proceed without asking someone.

Companies are now building tools around those decisions, including provenance, identity, verification, permissions, monitoring, and governance.

“Capability without trust simply increases the speed at which mistakes propagate,” Taub says.

That is what makes the events of April 7 more relevant now than when they happened. The false tariff claim was corrected quickly, but financial markets had already reacted before those corrections caught up.

Human traders have supervisors, risk limits, compliance rules, and ultimately people who are responsible for their decisions. An AI agent can be given versions of the first three, but there is no obvious equivalent for the person ultimately responsible.

If an AI agent reads the information, weighs the evidence, and makes the call, at what point does that decision still belong to a person? For now, that line is still being drawn.

Kalshi’s COO thinks prediction markets can beat the polls

1 month 1 week ago

Prediction markets incentivize truth. Everything else incentivizes clickbait. That’s the core argument Luana Lopes Lara makes for Kalshi, the prediction market platform she co-founded that has gone from $5 billion to $22 billion in valuation in under a year and is now being sued by New York’s attorney general for $36 billion. Lara, the company’s chief operating officer, makes the case that Kalshi is fundamentally different from DraftKings—even though 75% of its volume comes from sports—and shares what Kalshi’s data actually says about the midterms.

This is an abridged transcript of an interview from Rapid Response, hosted by former Fast Company editor-in-chief Robert Safian. From the team behind the Masters of Scale podcast, Rapid Response features candid conversations with today’s top business leaders navigating real-time challenges. Subscribe to Rapid Response wherever you get your podcasts to ensure you never miss an episode.

Kalshi launched in 2021. Last year at this time, Kalshi was valued at $5 billion. I think by early this year, it was up to $22 billion. That scale at that speed—what makes that possible? Is there luck in it?

It’s a great question because actually, though we launched in 2018, right, in a lot of ways it looks like an overnight success. Like it was all up two years ago, and we just started growing a lot around the election. But it was the result of eight years of work. When we started the company, it was very important for us to be legal and regulated from the start, so it took us four years before we could launch the product, launch anything really, or have any users. We worked with the federal government to figure out how to bring prediction markets to the U.S. in a safe and regulated way.

A lot of things helped us grow this much now, but I think it’s the compounded effort that the team has put in for so many years on the tech, on the users, and on talking to them. Then we won the lawsuit against the CFTC [Commodity Futures Trading Commission] to be able to bring a lot more markets to Kalshi. When that happened, the product was ready to really grow and go from there. So I think it’s a mix of both. We were very prepared when our time came.

With that kind of hockey-stick growth, do you have to pinch yourself? Is this totally real? Is there anything about that pace that scares you?

I actually would say it’s a very good thing that it happened so fast because, in a lot of ways, we keep the mentality of being very early stage. I think when companies are compounding at a very normal rate, it’s easier to start thinking, “Oh, I’m a bigger company. I need to hire more people,” and you can start making a lot of mistakes. It can take a long time for you to realize you’re making them. For us, because we grew so fast, our mentality and the way we look at the company haven’t changed as fast.

Because of that, we’re able to operate with far fewer people. We’ve just had to keep going and building the product as fast as we could: early-stage team, early-stage mentality, and very intense work. Keeping speed is the most important thing for startups. You’re definitely right that sometimes we look at the numbers and, two years ago, before the election, we were making way less than 10 million dollars a year.

Now, in a day, we transact way more than we used to in a year just two years ago. It is crazy, the numbers we’re talking about, and we’re very grateful for where we are. But we really try to keep the mentality that we’re still underdogs, and we still have a lot to prove and a lot to grow.

The success you’ve had has put a bull’s-eye on your back. States are coming after you for being an unlicensed gambling operation. A federal appeals court just ruled that Ohio and Tennessee can regulate Kalshi through their gambling laws. Is that kind of an existential threat? New York alone is suing you for $36 billion . . . the state where you’re headquartered.

We are very confident in our legal analysis. Of course, as you said, the appeals court went against us, but we also won the 3rd Circuit. Each of these lawsuits has a different legal thesis. The more important part, if you take a step back, is that the mechanics of how Kalshi operates and how a sportsbook operates are completely different, right? And that’s why they are regulated in different ways. We are federally regulated. We are an exchange, which means you trade against someone else. We don’t set the price. We don’t set the odds. We don’t trade against the users. The users are trading against each other, and we take a transaction fee.

What matters most here is liquidity and making sure we have national liquidity to build on. Imagine if you had the New York Stock Exchange, but you could only buy stocks on the New York Stock Exchange if you were in New York. The prices would be significantly worse. It would not be a liquid market. It would just be worse for every participant, and the market wouldn’t work well.

A sportsbook, on the other hand, operates completely differently. Because we also don’t trade against our users, we don’t make money when users lose. A sportsbook is completely different. Their revenue is equal to customer losses. The more the customers lose, the more money they make. For us, it’s not the same. The incentive is not to make people lose because we don’t make money when people lose.

Because of that as well, we don’t cap our winners. If you go to a sportsbook or casino and start making money, they’ll make sure you cannot participate anymore. We want winners. We want people to come and bring price, and we want price competition. In a sportsbook, there’s no price competition. The sportsbook has a monopoly on the price, and they’re going to put their margins on top because they’re having a bad month, so they make the prices a little worse or whatever. Because of that, they are fundamentally different mechanics and fundamentally different products, and they need to be regulated in different ways, which is how the federal regulation for exchanges developed.

We’re growing a lot because an exchange is a fairer, more accessible, and more transparent way to trade. You can see all the prices. You can see the competition and the order book in real time, and that’s why users like it so much. I think it’s fair that consumers, at the end of the day, pick what’s better for them.

Election polling has become kind of unreliable. What does Kalshi’s current data say about the U.S. midterms? Is that 70% accuracy? Or at what point does it start to move toward 90%?

A poll is top-down, right? It’s some editorial board or someone doing a poll, trying to aggregate the information and just tell people, “This is the number” or “This is the forecast” or “This is what’s going to happen.” But prediction markets are bottom-up, right? We want as many people as possible to do as much research as possible and bring that information to the market. So it is kind of an aggregation of what millions of people are thinking and doing, and I think it’s one of the first times that you really see information that’s actually led by people versus the elites just coming and saying, “This is what’s going to happen.”

With our big markets, like who’s going to take control of the Senate or control of the House, I think it’s as accurate as you’re ever going to get. The other thing we always have to talk about is that probabilities are not certainties, right? When something happens 1% of the time, it doesn’t mean it will never happen. It means that one out of 100 times, it will happen.

An election that’s at a 60% chance of someone winning means there’s still a real chance the other person wins. If I told you if you walk outside right now, there’s a 40% chance you’ll get hit by a bus, you’re not going to walk outside because 40% is pretty high. It’s the same thing. Forty percent does not mean that the person, the underdog, is never going to win. I think that’s kind of a challenge we have on the educational front, which is explaining to people that, different from polling, this is not the same thing.

Polling might say someone is 10 points ahead. In the market, that would probably mean over a 90% chance of someone winning, because they’re very different things. They measure different things, and we need to look at them as probabilities.

‘We’re actually an intelligence organization now’: Inside USA Today’s AI-fueled Palantir deal

1 month 2 weeks ago

When the nation’s largest newspaper chain announced a deal with Palantir in August, USA Today Co. CEO Mike Reed framed it as a necessary modernization step for a print-heavy publisher. The company, which operates USA Today and more than 200 local publications, still generates $1.1 billion in annual print revenue, but faces the same circulation declines and changing digital habits battering the rest of the industry.

Reed laid out the AI-fueled vision at a Palantir conference last month. The company wants to consolidate years of fragmented reader data into a single system and use AI to personalize newsletters, target affiliate links, improve content discovery, and turn anonymous visitors into registered subscribers. Demoing the platform with an “anonymized” user named “Marcus,” Reed described a continuous feedback loop built around reader behavior. “We’re actually an intelligence organization now,” he said, “because of the shift to online and the data we have on consumers.”

Last month, CEO Mike Reed demoed Palantir’s platform analyzing an example “anonymized” reader [Photo: Palantir]

The reaction from the company’s journalists was swift and, perhaps, unsurprising. The deal, announced during an August investor call and first reported by Nieman Lab, prompted a revolt across 31 unionized newsrooms represented by the NewsGuild-CWA. In a joint letter calling for the contract to be terminated, staffers argued that Palantir’s work with U.S. Immigration and Customs Enforcement creates a conflict for reporters covering immigration enforcement and protests, while raising questions about reader data and public trust. In addition to a booming commercial business, Palantir—fronted by its chest-thumping CEO Alex Karp—also holds billions of dollars in federal contracts, including for the Defense Department’s flagship data fusion platform. “Every problem we have with this partnership really comes down to trust,” Mike Davis, an investigative reporter at the Asbury Park Press, told Straight Arrow News. 

That skepticism comes after years of cuts. Since the company formerly known as Gannett merged with GateHouse Media in 2019, it has gone through repeated rounds of layoffs, essentially halving its workforce by 2024, even as CEO compensation rose amid flat or declining median employee pay. The company had roughly 7,500 U.S. employees last year, about 15% of them unionized, according to its latest SEC filing. Leadership has cast the restructuring as a response to falling readership and new forms of distribution and discovery, including AI feeds and chatbots. Much the same logic now underpins its AI push and its alliance with one of the world’s most controversial technology companies.

USA Today isn’t the only media organization to partner with Palantir, or face scrutiny over surveillance. Thomson Reuters has long supplied data used in federal immigration surveillance tools. Marketing firms Stagwell and Zeta Global have worked with Palantir on ad targeting. Fox News Digital used Palantir to build an internal platform that supports much of its article production, while Axel Springer, owner of Politico and Business Insider and many other outlets, has partnered with the company to better understand audience behavior.

Last month, Axel Springer also gave Palantir’s cofounder and chairman Peter Thiel its 2026 Axel Springer Award, recognizing him “for his transformative impact on the digital age.” A decade earlier, Thiel, who also founded PayPal, secretly bankrolled Hulk Hogan’s lawsuit against Gawker Media, contributing to the publication’s bankruptcy and demise. Some USA Today staffers see that episode as part of a broader pattern of Thiel-linked hostility toward the press, and another reason to ditch the contract.

Reporters have also lately raised concerns about the impact of AI on their work and their livelihoods. USA Today-owned newsrooms are using AI tools for research and investigative tasks, like filing public records requests, and to generate headlines and automate parts of digital publishing. After an abortive experiment in AI-written sports stories starting in 2023, the company the following year rolled out AI-generated “key points” that sit at the top of stories. USA Today Co. policy says that if any “additional AI-generated content is published, journalists must include AI disclosure language for transparency and “must be verified by a human for accuracy prior to publication.” Last year, WGBH reported that several of the company’s Boston-area news publications were using an AI tool called Espresso that is “designed to draft polished articles from community announcements.”

At the same time, the company is feeding its journalism into generative AI products, including its DeeperDive answer engine and through distribution and licensing deals with Meta, Amazon, and Microsoft. (It has also sued Google, along with the Daily Mail and a class of about 5,000 other publishers who allege the search giant overcharged them through its advertising technology platform, and on Thursday, it sued OpenAI for allegedly stealing its content to train its large language models. ) As Reed put it on the investor call: “We recognize that we have to create and format content for humans and for machines.”

In an interview last month, Reed, Media President Kristin Roberts, and communications executive Lark-Marie Anton spoke about the rationale for the Palantir deal, how AI is actually being used in the newsroom, and the union friction it’s provoked. Contrary to what they called falsehoods by union leaders, they said Palantir’s software, assembled with the help of a team of embedded Palantir engineers, works only with governed, “anonymized” data, and said the company has no direct access to newsroom systems, reporter notes, or subscriber records. Instead of threatening journalists’ jobs, Reed said AI tools could make the newsroom and business more productive and profitable, and, he hoped, would allow the company to hire more journalists. 

Following the publication of this interview, NewsGuild President Jon Schleuss responded in a statement: “USA Today Co. executives’ animosity toward their workers and unions is nothing new. The NewsGuild, which represents over 800 unionized USA Today Co. workers, fully stands behind the statement workers released when this deal with Palantir was first announced.”

This conversation has been lightly edited for clarity. Anton provided some additional written answers shortly before she departed the company on October 1.

How did USA Today Co. come to work with Palantir? What was the spark for this relationship? 

Mike Reed: I think that just being connoisseurs of the media, we knew some of the problems we had with having a lot of data, but also having multiple systems and not being able to have our data cohesive, in one place, and being able to learn from it and have recommendations from it. And so understanding that problem and talking about it internally and thinking, well, this is going to take years and it’s going to be expensive to maybe get ourselves to a place where we can use this in a real business-oriented way. And watching a lot of interviews with CEOs who were partners with Palantir saying that with Palantir, they did in a matter of a month or two things that might have taken them years to do. And I just thought, we need to talk to them, because whether it’s cheap, expensive, whatever it is, we have a business issue to contend with, and trying to build something ourselves that might take three or four years—and we don’t even know if we’re going to get it right and it might cost a lot of money. Or, we could partner with somebody like Palantir, and be ready next month. 

And so that’s why we said let’s talk to them. And we liked what we heard from them, because it was obvious that they were going to be execution-oriented, not just, we’re going to think about your problem and tell you what to do. They’re like, we’re going to actually do the work and help you execute. And that was really refreshing. So we said let’s do it.

Did you consider other companies?

Anton: We evaluated the landscape broadly because our challenge is not simply analytics, personalization, or AI. It’s the need to connect a very large, complex set of audience, advertising, subscription, commerce, and distribution data into a common operating environment. What stood out about Palantir was its ability to create what it calls an ‘ontology’ layer—essentially a unified model that allows different datasets and systems to work together and be acted upon in real time. That capability was particularly relevant to a company like ours because we already have significant first-party data and strong audience reach. The challenge isn’t collecting more data, but making better use of the data we already have. As Mike has said internally, Palantir is operationally more sophisticated in this area than any vendor we reviewed.

CEO Mike Reed [Photo: Courtesy USA Today]

Is there a metric you’re focused on here, or is there something that you think so far has proven to be valuable?

Reed: Not yet is the short answer. We started working with Palantir in June. So we really only have a couple months under our belt, and most of that time has been them really organizing our data, trying to take all this data from disparate systems and make sense of it in a manner where they can use it, they can provide the business outcomes we’re looking for. When we started to engage with them, we came up with a list of, here’s the business outcomes we’d like from you. Here are the problems we’re solving for. And then they worked with our data to try to figure out exactly what the solutions would be. So that’s really where the first couple months have been. But what might have taken us a couple of years, they figured out shortly, and will be in the market here as we go into the fourth quarter and start to, hopefully, have some positive learnings.

What are some of those outcomes you’re aiming for?

Reed: One is, how do we engage longer with consumers on our platform? We don’t want consumers to come for one story and leave. So how do we make the experience more enjoyable and more personal? Our audience is massive. A lot of times they’re there for a couple minutes and then they’re gone. And so how do we keep them there, how do we engage with them and how do we get them back more often?

And the second thing we’re trying to solve for is, we have a lot of users that come to our platform whom we don’t know anything about. They’re not registered, they’re not subscribers. And so while we can still derive some value from them, it’s beneficial. If we can create known users out of unknown users, the value we can create from our relationship with that consumer goes up exponentially. 

What is Palantir being paid, and how does that compare to what an in-house build would have cost? What’s the length and exclusivity of the contract, and is there an off-ramp if the results you’re after don’t materialize?

Anton: I will share that we approached this with the same financial discipline we apply to any strategic technology investment. We evaluated the expected return, the speed to value, the opportunity cost of building internally, and the resources required to build, support, and maintain a comparable platform ourselves. And we have been focused from the outset on flexibility and long-term independence. One of the key workstreams has been ensuring knowledge transfer and building internal capability so that we are not dependent on outside engineers forever. We have also evaluated off-ramp scenarios and continuity plans as part of the process: If we don’t see measurable business value, we will reassess just as we would with any strategic investment.

How do you think about USA Today’s goals as part of a larger structural shift that’s happening across news media—or a shift that you think will need to happen?

Reed: I think a big structural shift for news media like ours is that, when we think about content creation, it can’t just be hard news. We have to create a lot of sports content and entertainment and games and puzzles. If we’re going to engage with consumers for longer periods of time and more often, we have to have a broader content strategy than just news. So that’s a big structural change for an industry that’s really created news historically, and not really focused on a lot of that other stuff. And that other stuff’s really important now.

And the second thing is listening to the data, in terms of your content creation strategy. And if you think about it, not that long ago, your newsroom would meet in the morning around a table and say, ‘What are we going to cover today?’ And you would go cover it, and that’s the news that would be reported. And our people sitting around the newsroom, they don’t have bad ideas, but why not listen to the consumers? What do they want covered? And so looking at the data and understanding what’s resonating with consumers, what are they focusing on and doing more of that and looking at the data and saying, ‘They’re not paying any attention to this, maybe we should do less of that or none of that.’ That will allow us, I think, to create more content that’s engaging to consumers. 

Don’t mistake that to mean that we would abandon the social mission side of our business, which is to cover the communities we operate in, cover the country, do investigative work and do the things that are going to protect citizens and people who live in our communities. Of course we’re going to do that work, but there’s hard news that we cover that’s not necessarily aligned with that mission that people don’t care about. So let’s do less of it. 

How are people in the newsrooms responding to these changes? What questions do they have, and how are you responding to them?

Reed: I think that a lot of folks in our newsroom are covered by the NewsGuild, and I think the NewsGuild gives them a lot of bad information that leads them to decision-making that is not the best. And so I think the News Guild is doing a disservice to the people, its own members who are our employees and that are part of this cohort of journalists that have to make these changes. And I think that makes it harder, and I think we’re having more success in newsrooms that are not unionized, with this new world, than those that are unionized. 

Kristin Roberts, USA Today’s Media President [Credit: USA Today]

Kristin Roberts: Guild leadership is saying things that are factually untrue. They prey on fear. No matter that I will appear in front of any newsroom for a standup in which they ask me anything for an hour and a half at a time, and I explain to them all of the things. Every single journalist in the organization has my cell phone number and they can call and say, I just heard this from the Guild rep. Is this true? And I’ll tell them it isn’t. I’ll give you an example that’s incredibly real for us. Palantir doesn’t work inside of any USA Today Company system. They don’t have access to any of the notes our reporters take. They’re not in the Microsoft suite of tools. They don’t have information directly about our consumers or our staff. I said this to all of the staff, took their questions, and answered this precisely. They don’t work in our systems. We feed them highly controlled data. We’re in their shit. They’re not in ours. 

Reed: Two other untruths before we move on. There’s this fear—and I think the Guild plants this as well as outside cynics—is that we’re going to replace journalists with AI. That couldn’t be further from the truth. We’re hoping that our journalists can do more of the work they really want to do. And as we’re successful with our business strategy, we’re going to be able to hire more journalists. So this is not an exercise to replace journalists with AI at all. It’s hopefully going to allow us to add more journalists. Second is that I think the union plants fear that we’re going to have AI-created content, even though we say till we’re blue in the face we’re not. 

Reporters have also raised concerns about data protection. How do you ensure the data is protected when working with Palantir or any vendor? Are there independently verifiable guardrails that prevent sensitive data from being accessible to Palantir’s broader data environment?

Anton: This was one of the most important areas of diligence. Our position is straightforward: our data remains our data. We maintain ownership and control over it, and we require vendors to meet stringent security, privacy, and governance standards. Protecting personal information is fundamental to maintaining audience trust, and those requirements apply to Palantir just as they apply to any strategic technology partner. Data governance, access controls, ownership, security review, and legal protections were central considerations in negotiating the agreement and our ongoing working relationship.

Palantir burnished its reputation for years by staying clandestine, and some of its work is pretty controversial. How do you think folks, especially readers but also reporters and their sources, should think about this partnership in light of its shadowy reputation? Was there discussion about the reputational risks involved here? 

Anton: It’s a fair question. We recognize that Palantir has a public profile that prompts strong opinions. That’s part of the reason we’ve been deliberate about discussing not only the opportunities but also the governance, privacy, and trust considerations involved. For us, the key question wasn’t who else Palantir works with. The question was whether the partnership advances our mission while meeting our standards. Our journalism remains independent, and our editorial decisions are governed by our longstanding ethics and standards policies, and those policies do not change because we adopt a technology platform. 

We also work with other technology companies that serve governments, enterprises, and other complex organizations. We evaluate partners based on their capabilities, the strength of their controls, and whether the relationship aligns with our values and business objectives. Ultimately, readers should judge us by our journalism, our transparency, and how responsibly we steward their trust.

Are there other particular ways that you see AI helping inside the newsroom? And is that something that Palantir has been working on too? 

Roberts: A lot of the things that we have as reporters, over the course of my 30 years in journalism—I started as a news assistant at Reuters on the Wall Street desk—I have been asked to do more and more and more tech stuff, applying tags, applying quotes, inserting links, putting ads in different places. None of that has to happen anymore. Think of all the times with your editor where you were stumped on the right headline and you came up with 17, and you had to A/B test them. Now the system will give you A/B testing in real time. All of this stuff actually frees our mind to be creative, because you’re no longer spending 27 minutes searching for a lead video, wasting your time.

I’ll give you another example. In all of our entertainment stories and our sports stories, we do affiliate links to our partners. So if you’re reading a sports story, you want to buy the jersey for that team, our reporters and editors used to have to go to a file, find a relevant link, copy that link, put it in. All of that is now automated, and that’s a Palantir project. 

Can Palantir’s software help with social media and promotion? Can it help with ad targeting? Can it help reporters directly with their reporting, or could it eventually? 

Anton: I think the potential applications extend well beyond those initial examples. On the audience side, the goal is to better understand what a reader values and surface the most relevant experience at the right moment. For advertising, it can help connect audience behavior, content engagement, advertiser demand, and first-party data to improve targeting, relevance, inventory valuation, and monetization decisions. Hypothetically, could it support social strategy? Potentially. Any system that helps us better understand audience interests and content performance can inform distribution decisions across channels.

Again, hypothetically, could it help reporters directly? Perhaps over time. Other media organizations have used Palantir-powered tools to assist journalists with things like briefings, research support, and workflow tools. But the task at hand today is focusing on the audience and consumers to drive revenue and our business transformation.

How are your reporters using generative AI now?

Roberts: I think the vast majority of people in the company, and certainly in the newsrooms, are the ones adopting AI enthusiastically. So we’ve pushed all of our AI tools directly to our staff and say, Have some fun, figure this out. How can this make your job easier and better so that you can get back to the thing that you and I got into journalism to do, which is to break news and deliver scoops? All of our problems in journalism are solved if we get back to exclusive and distinctive, because AI answer engines can’t do that. 

For example, one of our news assistants in the UK created FOIAbot, and we have now been able to deploy that across the entire company. Say, I work in Naples, Florida, and I’m a local reporter. I want to see how many accidents happen at the intersection of First and Main. And so I go into FOIAbot and say, ‘I want to know how many accidents happened at First and Main.’ And FOIAbot identifies every single government agency that might have that information. With all of the formatting necessary for that particular government agency, it drafts the FOIA request, gives it to you, you sign it and it sends it. So we are able to send dozens of FOIAs a week. 

And the reporters are the ones who are creating these. So Kristin, sitting at an executive committee meeting saying, ‘How should my team use it?’ That’s not the way to do it. It’s to say, ‘Here’s the tool. You tell me, help me help you. You tell me how you can use it.’ And that’s what’s working for us, democratizing access to the tool. 

But with approved tools, where we have an enterprise agreement so that we are sure that the data is protected. You can’t go out freelancing with any tool you want. It’s got to be a tool that is given to us by a provider who has agreed to meet our very stringent standards around data protection. We have the highest standards for this. We’re a publicly traded company. And so any company that we do business with has to sign onto that, and we do not make exceptions when it comes to these AI. Then once they sign onto that, man, we push it to the people. We push it to the reporters and the editors and we say, let’s see what you can do with this. 

Given the concerns about AI’s impact on journalism, how do you think about changing the narrative within the organization? How do you get reporters more interested in trying it? 

Anton: We have town halls, where our own reporters, the ones that are the super users, share use cases. They talk about how they’re leveraging it. An example is our Taylor Swift reporter, Bryan West. You never know what Taylor’s going to do, so he may be out on the road covering something that’s not Taylor-related, and all of a sudden he’s going to go do the story, and he’s able to use AI. While he’s standing outside MSG, he’s using it to research stories that he’s already done to help him fact check something, or to create a listicle. So he loves it and he talks about it really enthusiastically. And that engenders trust amongst the rest of the team, to say, Bryan’s using it, then maybe I should try it for X, Y, Z. For [generating headlines], it took us a really long time to get everyone to say, This is actually great, that we have our own tool internally that gives you the key points at the top. 

So I think it’s a matter of, internally, people need to start to just trust the process. And they’re hearing their colleagues use it. I think it has to be sort of organic. You can’t force it. We’re not perfect, but we’re certainly trying to bring everyone along on the journey, and create trainings so they can understand how to use it.

Reed: We’re also transparent about what we won’t use it for. Then you have to practice what you preach. If we did use AI for something we said we weren’t going to, we’d have no credibility. Over a longer period of time, us being transparent about what we’re not going to use it for and then operating the business that way we think will be helpful. 

Are there things that you all like to use AI for, or that you find effective or that you’re hoping to use it for personally? 

Reed: I use it personally just every day to help me. I use Claude and it knows me better and better all the time. I use it mostly in a business context, but whether it’s helping me create the right narrative for an email or helping me develop solutions for another business issue or something like that. It saves me a lot of time if I want to research something. How does this apply? Or what does this really mean? Or how are other companies using this? I use it for that. I think it helps me become smarter a lot quicker. 

What sort of stance has the company taken on the use of its content for training and inference by the AI labs? 

Anton: Our view is that high-quality journalism and content has real value and should be compensated when it is used in AI products. We have pursued commercial arrangements that recognize the value of our journalism, provide attribution, and create new revenue opportunities that support news gathering. The [now expired] Perplexity partnership was an example of that approach. It was built around making trusted journalism available within AI experiences while ensuring recognition for the original reporting. More broadly, we support responsible collaboration between publishers and AI companies rather than assuming those relationships need to be adversarial. We want to make deals happen. But they need to be fair.

Digital publishers once had lucrative deals with Facebook and YouTube that evaporated once the platforms had the content and audience. 

Reed: Google does that all the time. They partner with you on business stuff and you build a business and as soon as they get it to where they want, they pull it up. 

Anton: We built on our own platform with a partner, Taboola. We were the first to implement on USAToday.com, our own AI answer engine that only pulls answers—so if you go to a story, you’re reading about September 11th, in there you can ask it to go deeper—it’s called DeeperDive—on the story. So you know that within our wall of USA Today, those answers are coming from a trusted source. I think that’s a way of us saying, ‘Hey consumer, stay here longer, ask all the questions you want and know that you can rely on those answers.’ I think that matters. It’s sort of a learning. We’re adapting to understand, we need to keep you here. And you need to trust us, and you need to keep coming back to us, not to a random Google [search] that’s going to give you an answer. It’s not going to drive you back to us for more information. 

When you look into the future—like five years—I wonder how you see newspapers, what their role is, how they fit into people’s daily lives. Does the newspaper as an idea change? Does the news business, does the media business have to change in some significant way or philosophically or structurally? 

Reed: I think that what we’re changing is that news is one form of content and that we’re a content business. And so we create news and other things, whether it’s entertainment or sports or games or helping them. 

But I think that in five years, I think the thing that’s continuing to evolve is we’re way more digital than we are currently. And in five years we’re going to be a lot more engaged with our consumers, because each one of them is going to feel like they’re getting what they need from us in a personal way. And so where we used to just push a set of stories into a market, we’re now going to be able to say, okay, we have this incredible pool of content. And for each user, pieces of that content are going to be relevant, and that’s what you’re going to be able to have. But we’re still going to be the news and media arm for consumers in local markets. But what’s evolving and changing is the personalized experience, and the fact that we’re a content business, and news is one form of content. There are many forms of content and we’re going to have to be good in all of this. 

And ultimately people want, I think, to trust the news they’re getting. And there’s a huge dearth of trust right now.

Reed: I think that’s a huge asset that plays in our favor five years out. Whatever social platform you’re on, there are so many things that are not real. And it’s hard to know what to believe. So much stuff looks real. And so that’s the one asset we have that’s becoming a bigger and bigger asset, that trusted news and information and content, whatever we create. We create now beyond text. We have audio, we have video, but whatever we create, there’s a level of trust with it. And I think that’s a real differentiator. 

And when I look out five years, I think one of the things that’s bringing consumers to us is not only that we’re unique from a local perspective, but that we’re a trusted source for all of the content you consume on our platform. I think that’s a differentiator five years from now. We’re going to become more valuable to the social media platforms too, because the owners of the social media platforms, they don’t want fake stuff. And so us creating content for those platforms I think is potentially not only a way for us to brand and bring consumers to us, but it’s a revenue stream from those platforms, because they’re going to want trusted content creators on their platform and we’re going to be high on that list. 

I hope there’s still a print edition in five years too.

Reed: I don’t know if the frequency will be as much as it is today, but I think there’ll be print in five years. I really do. Our print revenue is 1.1 billion, and it’s not going to go away tomorrow. It’s declining at eight to 10%. But we still have an enormous amount of readers around the country who want print. What’s interesting is our print subscribers automatically get a digital code too, so they can be a digital subscriber. Less than a third of our print subscribers use digital. They only want print. 

Correction: An earlier version of this story said USA Today Co. prohibits AI-generated stories. In fact, USA Today Co. policy permits AI-generated content with editorial oversight and disclosure.

Checked
4 minutes 57 seconds ago
Fast Company Technology
Fast Company inspires a new breed of innovative and creative thought leaders who are actively inventing the future of business.
Subscribe to Fast Company Technology feed