Skip to main content

Fast Company Technology

The midterms weren’t the rebuke of AI money many expected

1 week 6 days ago

The midterms are shaping up to be a record-breaking election for Democrats, who have flipped states including South Carolina and Texas. But the election has also shattered records for private political spending. By some estimates, super PAC spending exceeded $2 billion this cycle, more than 50% above previous election cycles.

One of the dominant narratives heading into the midterms was that Big Tech—and the artificial intelligence sector in particular—would spend hundreds of millions of dollars electing AI-friendly lawmakers, much as the cryptocurrency industry had done in previous elections.

Indeed, Leading the Future, the largest pro-AI super PAC network—backed by VC firm Andreessen Horowitz, OpenAI cofounder and president Greg Brockman, and Palantir and 8VC cofounder Joe Lonsdale—invited comparisons to the crypto super PAC network Fairshake, drawing on some of the same donors and political operatives. (Both networks operate through partisan affiliates: Leading the Future has Think Big for Democrats and American Mission for Republicans, while Fairshake has Protect Progress and Defend American Jobs, respectively.)

But as public sentiment toward AI soured and Democratic Party insiders warned candidates against accepting industry money, the narrative began to shift. A Fast Company analysis of election data suggests the midterms were hardly the rebuke of AI money that some had anticipated. More than 86% of the candidates backed by Leading the Future have won or are currently leading their House of Representatives or Senate races. That’s roughly on par with Fairshake, which supported 87 candidates this cycle, 77 of whom have won or are currently ahead.

The two industries’ political efforts also appear to be converging. Leading the Future and Fairshake backed 21 of the same candidates, including Republican Kevin Hern, who won his Senate race in Oklahoma. Hern also received support from the Anthropic-funded Public First, which has positioned itself as a counterweight to the industry’s other political spending operations.

Big Crypto outspent the AI industry but achieved similar results

While Big Crypto outspent the largest AI industry super PACs this election cycle and backed more candidates, the groups saw broadly similar results. Fairshake, the cryptocurrency super PAC funded by Coinbase, Ripple, and Andreessen Horowitz, spent about $154 million this cycle, compared with $34.8 million by Leading the Future and $24.4 million by Public First. Yet roughly comparable shares of the candidates each group supported won their elections.

Of the 12 Senate candidates backed by Leading the Future, nine won their races, compared with five of Fairshake’s eight candidates. Both Senate candidates supported by Public First—Hern in Oklahoma and Republican Pete Ricketts in Nebraska—also won. In the House, Leading the Future’s candidates won 31 of 38 races, compared with six of 11 for Public First and 63 of 79 for Fairshake.

Analysts had previously noted that Leading the Future appeared to be playing it safe, largely backing incumbents. But Fast Company’s analysis found that just over half of its 40 winning candidates were incumbents. The group also spent considerably more, on average, backing non-incumbents: roughly $511,000 per candidate, compared with $288,100 for incumbents.

Several of Leading the Future’s biggest investments paid off. Among its winning candidates were Sen.-elect Julia Letlow (R-LA), whom the group backed with $1.17 million; Rep.-elect Ben McAdams (D-UT), who received $1.12 million in support; and Rep. Ritchie Torres (D-NY), who received $1.16 million.

But heavy spending was no guarantee of success. Two of the 10 candidates who received the most funding from Leading the Future lost their races.

Arguably the AI industry’s biggest setback came in South Carolina, where Democrat Annie Andrews defeated Republican Darline Graham, sister of the late Sen. Lindsey Graham. Leading the Future spent about $2.3 million supporting Graham—more than it spent on any other candidate. Another pro-AI super PAC, Building the Future, poured an additional $2.5 million into the race, the only contest it has reported spending on. Yet Andrews, a former pediatrician, pulled off a historic victory in the Republican stronghold without financial backing from either AI or crypto super PACs.

Leading the Future also spent nearly $1.4 million on Democrat Jesse Jackson Jr., the disgraced former Illinois congressman who resigned amid a federal corruption investigation. He didn’t make it past his primary.

Public First suffered several high-profile defeats of its own. The group spent roughly $13 million backing New York Democrat Alex Bores, who lost his primary. Other losing candidates who received substantial support included Texas Republican Carlos De La Cruz Jr. ($1.5 million), Alabama Republican Rhett Marques ($960,000), and Texas Republican Alexandra Mealer ($905,000).

Fairshake likewise saw some of its biggest bets fail. The group spent $36.8 million supporting Ohio Sen. Jon Husted and another $19.7 million opposing his Democratic challenger, Sherrod Brown. Brown won. Fairshake’s second-largest investment, however, paid off: Alabama Republican Barry Moore won his Senate race after receiving $12.2 million in support.

The group also spent millions trying unsuccessfully to defeat several Democrats, including Illinois Senate candidate Juliana Stratton, whom it spent nearly $10 million opposing. Three House Democrats also prevailed despite Fairshake’s opposition: La Shawn Ford in Illinois ($2.5 million), Oliver Gilbert in Florida ($2.1 million), and Donavan McKinney in Michigan ($269,000).

AI Super PAC spending peaked during the primaries, then cooled heading into the general election

Weekly spending by the largest AI super PACs surged in June, coinciding with the closely watched Democratic primary of Alex Bores in New York. Bores, a New York Assembly member and former Palantir data scientist-turned-AI safety advocate, lost after Leading the Future poured $8 million into opposing his candidacy. The Anthropic-backed Public First, meanwhile, spent roughly $10 million of its $13 million investment on Bores that same month.

Bores lost to fellow Assembly member Micah Lasher, who went on to appear on the general election ballot in New York’s 12th Congressional District. Neither super PAC spent money supporting or opposing Lasher, who was victorious on Tuesday.

But AI money, especially toward the end of the election cycle, was dwarfed by Big Crypto.

In October, Fairshake spent $57.6 million, largely on advertising and media buys, compared with $4.1 million by Leading the Future and just over $1 million by Public First.

Industry observers have attributed that disparity, at least in part, to AI’s growing unpopularity across the political spectrum. A September Pew Research Center poll found that 56% of Democrats were concerned about the impact of AI, compared with 46% of Republicans—a relatively narrow partisan divide in an otherwise polarized political climate.

Even so, both Leading the Future and Public First backed more Republicans than Democrats. Leading the Future supported 34 Republicans and 16 Democrats, spending money to oppose just one candidate: Bores. Of the Republicans it backed, 28 won or are currently leading their races, compared with 15 of the 16 Democrats. Public First supported eight Republicans and five Democrats, while spending $1.4 million opposing three Republicans. Four of its five Democratic candidates won, compared with four of its eight Republicans.

Despite the prevailing view that Democrats see AI money as toxic, Big AI has still spent an outsize amount supporting Democratic candidates. On average, Leading the Future spent nearly twice as much per Democratic candidate than per Republican candidate. 

Public First’s spending followed a somewhat different pattern. Although it spent more overall supporting Democrats, largely because of its $13 million investment in Bores, its median spending per Republican candidate was higher.

That spending came despite a public backlash from within the Democratic Party. More than 100 Democratic operatives signed an open letter urging candidates to reject money from AI companies. Yet the industry’s political spending continued, often with little mention of AI itself. As NBC News reported, ads funded by AI groups rarely highlighted candidates’ positions on the technology.

There is also a convergence between the candidates that Big Crypto and Big AI support. Leading the Future and Fairshake backed 20 of the same candidates this cycle, including New York Democratic Rep. Ritchie Torres, whose campaign received $1.4 million from Fairshake and $1.1 million from Leading the Future. Alabama Republican Sen.-elect Barry Moore also benefited from both groups, receiving $12.2 million from Fairshake and roughly $330,000 from Leading the Future.

As public sentiment sours, the industry faces a wave of potential legislative and regulatory challenges. Democratic gains in Congress could mean greater scrutiny of AI companies, which have until now enjoyed powerful allies in the House, Senate, and White House. Proposals ranging from state and federal data center moratoriums to restrictions on the development of artificial superintelligence could help shape the industry’s future.

When the new Congress convenes, roughly one in 13 House members and a quarter of newly elected senators will have benefited from Leading the Future’s support. For a super PAC that didn’t exist two years ago, that’s a remarkable foothold. Whether it translates into political power is another question.

Anthropic’s AI development has doubled in speed despite slowdown calls, research group says

1 week 6 days ago

Anthropic is developing its artificial intelligence systems around twice as quickly as it was a year ago amid the CEO’s calls to slow down the technology’s pace of improvement, according to an analysis released Thursday by an independent monitoring group.

Anthropic invited researchers from the Berkeley, California-based nonprofit, Model Evaluation and Threat Research (METR), to review its internal data and interview staff as it prepares for its Wall Street debut with debate swirling over the potential for AI models to elude human control with catastrophic consequences.

The findings are significant because details about the technology—and its risks—have been provided primarily by AI companies, which have been encouraged by President Donald Trump’s administration to police themselves on matters of safety.

It was during the research group’s assessment period that Anthropic CEO Dario Amodei called for the AI industry to slow its work so safety measures can catch up. In a September 12 essay responding to mounting concerns about runaway AI, he urged the AI world to “slow the pace at which we improve the capabilities of AI models.”

Anthropic did not immediately comment on the report in response to inquiries.

The San Francisco-based company is planning to go public later this month in what could be the biggest initial public offering of stock ever.

AI is playing a bigger role in its own improvement

METR estimated that Anthropic is improving the capabilities of its models to do “well-specified tasks” around twice as fast as it was a year earlier. The nonprofit’s researchers said they believed much of the reason for the faster development is the help of AI itself.

Anthropic has said its Claude chatbot is increasingly helping the company develop newer, more advanced versions of itself.

METR has conducted evaluations on model capabilities for Anthropic, OpenAI, and other labs for years. The nonprofit is funded through donations and does not accept compensation for its work, but AI companies it has partnered with for evaluations have provided access and tokens, which METR uses for evaluations, research, and engineering.

One of its two reports was based on analysis of Anthropic’s internal data provided by the company over a period of about 11 weeks, from mid-August through late October, as well as employee interviews.

METR said its assessment focused on two questions: whether the growth rate of Anthropic’s internal capabilities has accelerated and how much of that was attributable to AI. The group said it was given “unprecedented access to internal data” on Anthropic’s AI research and development process.

Anthropic’s redactions to the report limited how METR could communicate more detailed estimates of acceleration, and there were some internal disagreements and uncertainty among METR’s team, but the group said it believed the company’s rate of improvement had doubled.

“Anthropic appears to have made as much progress on benchmarks in the past year as they did between October 2023 and October 2025,” according to the METR report.

METR also reviewed both a public risk report Anthropic published in August and a private version, as well as other non-public data, and the organization said the public report “overall understates the level of acceleration that was occurring.”

Assessors see little holding back AI’s progress

The report was peppered with caveats. METR said it could be overstating its findings, because the acceleration might only be limited to “well-specified tasks” that can be graded automatically or with a language model, instead of “messy” tasks that are harder to define. Or it could be that Anthropic understated its previous progress, making the latest number look bigger.

METR said it was able to “confidently” rule out the growth of computing power used in training AI models as the reason for the acceleration. Traditionally, the amount of computing power, known as compute, is directly related to how well an AI model performs. The researchers said their guess was that the improvement was down to using AI itself to improve AI.

A key concern, METR said, is that using AI to improve itself could start to compound the acceleration, which could be especially worrying given that researchers found there were no “bottlenecks” such as restrictions on computing power or data holding back Anthropic’s AI progress.

“We understand that AI is not yet able to make highly significant and fully autonomous contributions to the most important parts of the model-training process,” the report said. However, Anthropic researchers described significant “signs of life” implying that could happen soon, METR said.

Amodei touched on this issue in his September essay, saying AI advances have been driven primarily by the concept of recursive self-improvement, or what he defined as “AI’s growing ability to build the next generation of AI.”

“Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all,” Amodei wrote.

In that same essay, Amodei championed the idea of having external evaluators embedded within the leading AI labs and specifically highlighted METR as an appropriate third-party evaluator. OpenAI CEO Sam Altman said at the time that he also supported the idea of independent evaluators with employee-like access.

—By Kelvin Chan and Kaitlyn Huamani, Associated Press

7 ways the midterms may affect the AI boom

1 week 6 days 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.

There were 41 ballot initiatives limiting data centers; all but one are on track to pass

Voters in 10 states weighed in Tuesday on 41 local ballot measures related to data centers, Forbes reported. Ohio led with 21 measures, followed by Illinois with seven, Maine with four, and Kansas with three. Six other states—California, Florida, Nevada, New Jersey, North Dakota, and Wisconsin—had one each. Of the measures with reported results, all but one went against data center development, reflecting widespread opposition to the facilities.

In Ohio, voters overwhelmingly backed restrictions on data centers, including outright bans in several municipalities. More than 80% of voters in Ashville rejected an agreement involving two proposed data centers, though the vote does not itself halt construction. In Kansas, Emporia approved a ban on high-impact data centers, while Ottawa voted to prohibit facilities consuming more than 25 megawatts of power. Boulder City, Nevada, rejected a proposal to permit data centers in the Eldorado Valley Transfer Area. One notable exception came in Harwood, North Dakota, where voters rejected a measure that would have required additional review of certain industrial developments, including data centers.

Democrats say AI hearings and legislation are coming

“Two things will change next term,” Representative Ted Lieu of California, vice chair of the House Democratic Caucus, said at a press conference Wednesday. “We will hold public hearings on AI. We will pass legislation, reasonable guardrails, on AI.” Lieu criticized Republicans for failing to regulate the industry while they controlled Congress and said Democrats would also consider restrictions on new data center construction.

The odds of strong AI legislation “just shot up,” Shakeel Hashim wrote in Transformer. He argues that Democrats’ commanding congressional majorities would reduce the need for compromise. If the party wins 56 Senate seats—it currently has 54, with two races uncalled—it would need just four Republicans to overcome a filibuster. Hashim believes that support could be easier to secure as Republicans begin distancing themselves from Trump. Hashim predicts that Congress could within months pass legislation tougher than the bipartisan FRONTIER Act (cosponsored by Jay Obernolte, Republican of California, and Lori Trahan, Democrat of Massachusetts), including mandatory testing and approval of the most powerful artificial intelligence models and tighter limits on federal preemption of state laws.

Stocks rallied on the blue wave

Trump was supposed to be the president Wall Street and the AI industry wanted. Yet markets rallied on news that he had lost control of Congress. Global stocks initially fell as Tuesday’s midterm results trickled in, then rebounded as the scale of the Democratic sweep became clear, The New York Times reports. The S&P 500 closed 1.6% higher on Wednesday. Asian stocks initially tumbled amid fears that Democratic control of Congress would slow the AI buildout. Taiwan Semiconductor fell 1.5%, while SoftBank dropped 3.2%. But those concerns quickly faded. South Korea’s KOSPI reversed course to close up 0.6%, while European and U.S. stocks rallied. As the Times notes, investors have historically welcomed divided government, and the resolution of election uncertainty can matter more to markets than the outcome itself.

CoreWeave, a bellwether for the AI bubble, saw its bonds hit new lows

CoreWeave’s bonds tumbled Wednesday after Democratic candidates who campaigned on slowing data center construction won governorships, Bloomberg reports. Its unsecured debt due in 2031 traded as low as 79.5 cents on the dollar, among the worst performers in high-yield bonds, with its 2030 and 2032 obligations near distressed levels and yields above 15% (issuers must offer higher yields as confidence that the debt will be repaid erodes). In Texas, Gina Hinojosa ousted Greg Abbott to become the state’s first Democratic governor in more than 30 years after repeatedly calling for an immediate moratorium on data centers. Two governors-elect, Amy Acton in Ohio and Keisha Lance Bottoms in Georgia, also ran on slowing construction.

CoreWeave’s stock was already falling. CoreWeave shares had declined approximately 40% over the preceding month as investors grew concerned about the company’s debt, according to Morningstar data. Its bond yields had reached 15% or higher in October, well before the election. The company reports third-quarter results on November 10. Among the questions facing investors is whether rising bond yields have increased CoreWeave’s cost of borrowing, and whether it is bringing new data center capacity online on schedule.

The last constraint on the AI buildout is political

Ben Thompson, a business and tech analyst, argued in his newsletter Stratechery that the main constraint on building more AI data centers will soon have little to do with physical resources. “First we were short on compute, but that ramped; now we are short on memory, but that will ramp; we were coming close to being short on power, but that will be rectified,” Thompson wrote. “The final constraint is political, and that may be the most imposing constraint of all.”

Thompson refers to the public angst over new data center builds that was clearly expressed in the midterms, and he doesn’t expect that angst to lift anytime soon. “This isn’t a partisan issue, where the pendulum will swing back in two years. Both parties see opposition to data centers as a winning cause,” he says, pointing to local officials recalled for approving projects in their own communities. 

A scarcity of compute power could have some unintended, and positive, effects. The fierce competition among frontier AI labs has prevented them from moderating the pace of their research on new models, which, evidence shows, may be outpacing the development of effective safety guardrails. A slowdown might also reduce the terrific amounts of money the labs must spend to reserve computing power in the future, which could improve their overall financial position.

Elon Musk spent $474 million and still lost Texas

Elon Musk poured roughly $474 million into the midterms—nearly twice what he spent helping elect Trump in 2024—and had little to show for it. Of the 106 congressional candidates he backed, 44 lost, including Ken Paxton in Texas. Musk also backed the losing gubernatorial candidates Greg Abbott in Texas and Vivek Ramaswamy in Ohio. Just 18 of his congressional candidates had won as of Wednesday, all in seats Republicans were already expected to hold, while 44 races remained uncalled. Musk spent $212 million backing Paxton alone, only to see him lose to James Talarico, the Democrat, by 12 points.

The defeats could have real consequences for Musk’s business empire. Roughly a fifth of SpaceX’s $18.7 billion in 2025 revenue came from federal agencies, and Democrats will soon chair the congressional committees overseeing some of its biggest government customers. Democrats have discussed investigating DOGE, federal contracts involving SpaceX and Tesla, and Starshield’s role in Ukraine. As the Democratic wins piled up Tuesday night, Representative Robert Garcia of California, the incoming House Oversight Committee chair, tweeted simply “LAWYER UP”.

Meanwhile, in Texas, Hinojosa has proposed a moratorium on new data centers and will gain influence over state incentives and the commission that could decide whether to sell Boca Chica State Park to SpaceX. Investors appeared rattled: SpaceX shares fell 3.8% on Wednesday, while Tesla dropped 3.3%.

The AI industry’s biggest donors mostly lost

Democrats turned frustration with the tech industry into a midterm advantage, Politico reported, and, like Musk, Silicon Valley’s biggest donors have little to show for their GOP campaign donations. Marc Andreessen and Ben Horowitz contributed $116 million, mainly to Republicans. OpenAI’s president, Greg Brockman and his wife, Anna, put $50 million primarily into GOP campaigns.

Candidates ran against data centers and won. Much of Silicon Valley’s donation money went toward electing data center-friendly Republicans in Texas and Ohio, states where new facilities are planned. Hinojosa, in Texas, and Sherrod Brown, Ohio’s senator-elect, both ousted Republicans on voter unease about the facilities. Rebecca Cooke used the same playbook in Wisconsin, as did several Pennsylvania Democrats: Paige Cognetti, Bob Brooks, and Bob Harvie. Manny Rutinel, congressman-elect in Colorado’s 8th, made his record on state AI laws central to his campaign. “Allying with the worst parts of the AI industry is just kryptonite for a campaign, and the results tonight double down on that,” said Alex Bores, the New York state lawmaker who got buried by Big Tech money in a U.S. House primary this spring.

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 here.

Bluesky was right all along and will accept your apology any time

1 week 6 days ago

Bluesky is a niche Twitter-clone frequently accused of being an echo chamber for progressives. After the blue tsunami of this week’s midterms, however, its inhabitants don’t seem to be echoing themselves so much as the vast majority of American voters.

The sheer scale of Democrats’ victory is staggering. They won control of both chambers of the U.S. Congress, flipping Senate seats in seven states Trump carried in 2024—including Republican stronghold Texas, which now also has a Democratic governor and attorney general. All in all, this year’s midterms represent the biggest electoral shift in state-level politics to hit the U.S. in generations. They also represent a Texas-sized validation of the worldview Bluesky has become notorious for over the past two years.

Amusing how if you read X, Facebook, most of NYTimes and The Atlantic, all of The Argument, and so forth, yesterday was a huge surprise and you have to reset everything from the past two years.But if you joined BlueSky in November 2024 and clicked on any politics-related starter pack, you're good.

— Max Kennerly (@maxkennerly.bsky.social) 2026-11-04T11:27:40.179Z

BLUESKY WAS RIGHT ALL ALONG

— John Scalzi (@scalzi.com) 2026-11-04T04:19:19.671Z

Although it officially launched in the spring of 2023, Bluesky might as well have been born on election eve in 2024. That was the moment millions of disillusioned Twitter users decamped from a website owned by the largest single contributor to Trump’s reelection. Bluesky quickly became a haven for Twitter expats and grew at a frenetic pace. It went from 9 million users in September 2024 to 15 million by mid-November, eventually reaching its current 47 million.

Because many of Bluesky’s users arrived out of principled opposition to X, formerly Twitter, the two sites have arguably existed in opposition to each other. Both portray their counterpart as out of touch, with Bluesky alleging Elon Musk’s algorithm tweaks have skewed X’s political vortex too far to the right, and X suggesting Bluesky is a bubble of liberals who can tolerate anything except a difference in opinion.

The latter idea seemed to largely prevail. 

In 2025, Washington Post columnists, tech world figures like Mark Cuban, and myriad other pundits dismissed Bluesky as smug, self-righteous, and irrelevant. It was the fantasy sandbox where progressives went to play among themselves, protecting reality dwellers from having to hear their drivel. (“Like a containment dome over a nuclear reactor,” as the writer Josh Barro put it.)

So, how well did Bluesky’s users stay in touch with reality between elections, compared with X’s exponentially larger population? Consider the post-midterms perspective.

Rather than accept the narrative that 2024 represented a rejection of “wokeness” and a political realignment toward Trump, the manifestly trans-friendly Bluesky suggested the election was more a referendum on postpandemic inflation and global anti-incumbent bias. As an emboldened Trump then proceeded to antagonize U.S. citizens and the rest of the world, while making billions for himself in crypto, Bluesky meticulously chronicled the state of play. It became the premier destination for seeing radicalized older white ladies, abysmally attended pro-ICE rallies, and complaints about the lack of mainstream news coverage around massive No Kings protests. Any random day would find endless posts with evidence that, per the motto of the political philosopher and Bluesky power user Olúfẹ́mi O. Táíwò, “We are going to win.” 

— Andrew Lawrence (@ndrew.bsky.social) 2026-09-08T01:17:36.844Z

Above all, Bluesky proved to be less a bubble than an expression of the multitude of ideological factions that had united under the umbrella of non-MAGAness . . . to argue with one another.

Meanwhile, X’s millions of liberal and progressive folks were often drowned out by an algorithm that seemed to favor right-wing voices and their preferred narratives. A bubble chart from Nate Silver, the political analyst, visually depicted the accounts with highest engagement rates back in April. Its many MAGA-red circles made it look like the blood-filled elevator from The Shining. As a May 2026 study from Northwestern University suggests, X’s algorithm rewards posts that provoke strong reactions, so the loudest and most emotionally satisfying political claims become disproportionately visible. That’s how we got viral fever dreams about James Talarico, the then-senatorial candidate, checking into a mental hospital, something that never in fact happened.

It should come as little surprise that many high-profile MAGA voices on X spent the days leading up to the election incorrectly predicting how it would turn out.

I nailed all my election predictions in 2024, not so much in 2022. Here are my 2026 calls with a reminder that all of you should get out and vote. Byron Donalds wins Florida Gov comfortably. Senate: Ken Paxton wins Texas, Mike Rogers wins Michigan, Jon Husted wins Ohio, Ashley…

— Clay Travis (@ClayTravis) November 1, 2026

Don’t believe the polls. There will be no blue wave. Once again, fake polling, designed, not to inform but to persuade & suppress turnout, are misleading voters. I know what I’m talking about. As a twice elected Dem governor and 3 time elected Dem Congressman, I never lost an…

— Rod Blagojevich (@realBlagojevich) October 6, 2026

Who looks out of touch and stuck in an echo chamber now?

While X may have been more plugged into the political discourse online for the past two years, Bluesky now appears better attuned to what’s been happening on the ground. Its users were right about pretty much everything. 

They were right that anti-trans ads would have little or no impact on the 2026 election, and that their impact on the previous election had been vastly overstated.

They were right that Graham Platner wasn’t the only Democrat who could beat Susan Collins in Maine, and that Abdul El-Sayed’s association with a controversial Twitch streamer relatively few voters had even heard of was not as big a deal as X users seemed to think. 

And they were right that fleeing X’s algorithm helped them view the world with clear eyes.

If Bluesky is a bubble, then it is currently a champagne bubble awash in celebration.

The festivities will likely last only a matter of hours, though, until infighting explodes around just how the Democrats should spend their newfound political capital.

Playful wants the masses to start vibe coding

1 week 6 days ago

Mike Harrington and Darrin Massena think that making an app with AI is still too hard.

While you can ask Claude or ChatGPT to build something for you, that demands at least some awareness of programming frameworks and hosting options. AI app builders like Replit and Cursor also aim to be professional-grade development tools for businesses, not something you’d use to quickly scratch a personal itch.

Harrington’s and Massena’s startup, Playful, wants to bring vibe coding to a much broader audience. With a handful of simple prompts, Playful will build web-based apps and games that you can push to your phone and share with friends. Users can create up to two apps for free and use them for up to 30 days, at which point the service requires a $10 per month subscription.

[Photo: Playful]

“It’s for an audience that is less technical, for an audience that until recently didn’t even understand that they could make apps,” Harrington says.

Playful isn’t the first company to try bringing AI app creation to mainstream audiences. The highest-profile example is Wabi, which raised $20 million last year to be the “YouTube of apps,” though it has since pivoted to be more of an app-powered personal AI agent (while raising another $55 million). The App Store is also awash in apps like Vibecode, Vibecoder, and VibeStudio, which look to varying degrees like they were made with AI themselves.

Harrington and Massena, who worked together on Microsoft Bob and later sold their photo editor startup Picnik to Google, are betting that a combination of social sharing features and their own expertise in software development will help Playful break through. Harrington also cofounded Valve Software, where he worked on the hit PC game Half-Life. The duo launched their startup in 2023 under the name Hatch, intending to offer a no-code platform for web apps, and pivoted to artificial intelligence after realizing how it would make app creation much more accessible.

“We’ve always gone after things that are consumer-oriented, and fun, and to a certain kind of audience,” Harrington says. “Picnik . . . grew from just a little photo editor to where we had 60 million monthly uniques, and about 85% of them were women. It’s just a market that people didn’t go after, and I think there’s a similar thing in apps.”

[Image: Playful] How Playful works

Creating an app in Playful starts with a simple question: “What kind of app do you want to make, and who’s it for?”

What happens next is where Playful distinguishes itself from a typical chatbot. Based on your original prompt, it will ask follow-up questions on things like the features you most care about and how you’ll use the app.

For instance, I asked Playful to build an audio transcription app that cleans up the text, identifies speakers, creates topic shortcuts to skip to, and lets you tap on the text to hear the corresponding audio. Playful in turn asked about recording scenarios—interviews, lectures, or personal voice notes—and whether I’d want to do things like access old interviews or export full transcripts.

“It helps you with the process because it understands a lot of things about apps, making apps, what might make sense, what’s popular in apps,” Massena says.

After a few minutes, Playful presents a smartphone-sized app that you can run through Playful’s own app or turn into a standalone icon on your home screen. The apps are web-based, so they work on any device, without distribution through any app store.

Behind the scenes, Harrington and Massena say that Playful uses a variety of large language models to fulfill users’ requests. Compared with using a more general-purpose AI tool, Harrington says Playful has more specific training on the process of creating apps and how to distill users’ requests. It also has built-in tools for things like data storage, and it can offer AI capabilities like summarization within the apps themselves. The ability to connect external data sources such as email and calendars is on the way.

“No one has to worry about ‘Where’s my stuff stored, what kind of database is needed for this?'” Massena says. “If you look at a lot of the other products out there . . . there’s still a lot of technical problems that you have to get comfortable with.”

That’s not to say a single prompt is always enough. With my attempt at a transcription tool, for instance, Playful’s app initially crashed out to a blank screen, and then tried to impose a 20-minute recording limit. After some working with Playful’s chatbot to solve those problems, it took some extra back-and-forth to get proper transcript formatting, and I gave up on trying to label speakers as it just never worked well.

Even so, the resulting app was helpful in reviewing my recorded interview for this story, and Harrington and Massena say they’ve been impressed with what early testers have come up with. One user, for instance, built an entire Excel-style spreadsheet editor for personal finance, while another built their own take on Capcut for AI video editing and captioning.

“It’s amazing what people have built out,” Massena says.

[Image: Playful] A new kind of app store

Playful’s apps aren’t just for individual use. Users can share their apps with others via a link or email invite, and they’ll automatically get updates when the creator pushes out new versions. Playful also supports sharing a “Copy” of the app—with separate data for each user—and “Remixing” so that others can modify the app as they wish.

Massena says the long-term vision is for those shared, editable apps to be the “new normal” for app distribution, versus the “take it or leave it” approach of conventional app stores.

“Why wouldn’t you go to a place where you could pull an app off the shelf, it was incredibly high-quality, full-featured, but that you could change any way you want to meet your needs?” he says. “That’s going to become the new normal for apps.”

The only point of friction in this argument is Playful’s business model, which strictly limits what users can do without a $10 per month subscription. After 30 days, Playful puts free users’ apps to sleep, and it bars them from creating more than two apps total.

Massena says Playful initially experimented with a different approach, in which users received a certain number of “credits” to use as they wished. But that just led to anxiety from testers around how many credits they were using. Even without a subscription, Playful does want to allow users to try other users’ apps, at least to some extent.

“Maybe you have a new inspiration, something there you want to try out, we’ll make sure that that’s always possible for people,” Massena says.

Whether or not Playful succeeds, it illustrates the uncharted territory where we’re heading with software. The iOS App Store is already seeing a boom in submissions, and Microsoft’s GitHub has been drastically increasing capacity due to a sixfold increase in code commits. Last month, the developer Brandon Thomas declared that “software is over” after using AI to spit out a series of lookalikes for popular apps like Adobe Photoshop and Microsoft Word.

What does the future of software look like when anyone can build a free version of an app they like that also better suits their needs? It’s unclear, but Massena and Harrington hope to be at the forefront of the answer.

“Software companies are going to have to be more responsive to people, or it’ll be much easier to find alternatives,” Massena says. “We’re building a platform for that new future.”

From Texas to Michigan, voters are turning against Big Tech’s AI buildout

1 week 6 days ago

Data centers were on the ballot this election, and they lost.

While results from the midterm elections are still coming in, as of November 4, 38 out of 39 restrictions on data centers had passed—from temporary moratoriums on new construction to full bans. And multiple candidates who made opposition to data centers a key part of their campaign also won.

Take the example of William Lawrence, the Democratic candidate for the U.S. House of Representatives in Michigan’s purple 7th Congressional District, which Trump narrowly won in 2024. Lawrence, a progressive community organizer and cofounder of the Sunrise Movement, was new to politics and faced a challenge running against the Trump-backed incumbent, Tom Barrett. In the primary, facing two establishment Democrats who had raised more money, Lawrence won in part because he vowed to fight data centers.

“He found an issue that connected with local residents across partisanship,” says Matt Grossman, director of the Institute for Public Policy and Social Research at Michigan State University. Recent polls say that seven in 10 Americans don’t want data centers to be built in their communities, for reasons ranging from potential impacts on electric bills to noise and water use.

William Lawrence, a Democratic congressional candidate in Michigan, speaks at his election night event in Lansing. He defeated a Republican opponent. [Photo: Bill Pugliano/Getty Images]

In his first ad, Lawrence stood at a farm and said, “Tech billionaires want to turn this field into a massive data center that’ll jack up our energy bills and ruin our home values.” After going to local planning meetings to watch data center developers pitch projects, he said that it was an “abuse of power” in communities that didn’t have the resources to fight back. He also called for other controls on technology, including a pause on development of advanced artificial intelligence and mandatory safety requirements for AI.  

Barrett, his Republican opponent, had supported data centers in the past. That included a vote for a version of the One Big Beautiful Bill Act that would have stopped states’ ability to regulate AI, and likely also data centers, for a decade. (The provision was later taken out of the bill). The OBBBA includes tax breaks for companies building out AI infrastructure. Barrett also voted for the SPEED Act, which accelerates data center construction. As backlash to data centers grew, Barrett argued that communities should have local control over whether projects get built, but the message fell flat. Lawrence won.

Barrett was unpopular for other reasons—including the fact that he’d never held a local town hall—and it’s not clear that data centers were the thing that made voters push him out, versus anger about the Iran war, tariffs’ impact on the Michigan economy, and other factors. It’s probably too early to say that data centers as an issue swung a race, says Daniel Schiff, an associate professor of technology policy at Purdue University. But the election results affirm the opposition.

“If you had seen elections flip the other way, you would have said, oh, wow, this data center stuff actually isn’t that much of a concern for the public compared to other things,” Schiff says. That didn’t happen, and Democrats in other states who campaigned against data centers also won.

Gina Hinojosa takes the stage as she is announced as the next governor of Texas at her watch party in Austin on Tuesday, November 3, 2026. [Photo: Mikala Compton/The Austin American-Statesman/Getty Images]

In Texas, Gina Hinojosa became first Democrat elected governor since 1990, beating the Republican incumbent Greg Abbott, after calling for a data center moratorium. (Abbott had called Texas the “epicenter of AI development,” before pivoting to pause new data centers; Hinojosa said she would go further.)

In Georgia, Keisha Lance Bottoms, a Democrat, also won a race for governor after campaigning for a statewide moratorium on data centers. In New Mexico, the Democrat Deb Haaland pledged to implement a statewide moratorium and won the governor’s race. In Ohio, the Democrat Amy Acton called for a conditional moratorium and beat Vivek Ramaswamy in the governor’s race. Multiple candidates for Congress won on anti-data center platforms, like Sherrod Brown, who won a Senate seat after calling his Republican opponent the “face of data centers” in Ohio.

Opposition to data centers arguably was part of the reason that other races were much closer than expected—like in a race in southeast Georgia, where residents have opposed a new OpenAI data center. Trump won the district by nearly 50 points in the last presidential election, but the Republican candidate for Congress this year won by fewer than 3 points.

Keisha Lance Bottoms speaks during an election night watch party on November 3, 2026, in Atlanta. [Photo: Prince Williams/WireImage/Getty Images]

The question now is what public sentiment about data centers actually means for the tech industry’s plans for AI. It’s not clear that opposition to data centers is as much about AI as it is about issues like electricity prices and the climate impacts of soaring energy use. Those impacts aren’t baked in; it is possible to build data centers that run on clean energy paid for by tech companies, for example. But the skepticism about data centers is now so ingrained that it’s likely more moratoriums will pass.

That could mean that data center developers end up flocking to the few places left that might not fight them. “They’ll say, OK, we need to do this more quietly . . . we need to go to the communities that don’t speak back,” Schiff says. “That would be a plausible and unfortunate consequence.”

Tech companies are already struggling to access power for data centers, and as data centers themselves keep getting harder to build, it could also meaningfully slow down the spread of AI. Anthropic’s IPO filing said that the backlash against data centers could limit its ability to grow.

The tech industry has spent billions betting that it can build AI infrastructure fast enough to stay ahead of demand. But the midterms suggest that it may not be able to build fast enough to stay ahead of public opposition.

Meta paid billions, but largely escaped accountability for Cambridge Analytica. Here’s why

1 week 6 days ago

“The harm in this matter is difficult to overstate,” wrote a New Mexico judge who ordered tech giant Meta to pay one of the largest data privacy case penalties in American history. The $25 billion fine came in a case about how Trump campaign firm Cambridge Analytica deceptively used data from Facebook users without their permission to create psychographic profiles to target voters in the lead-up to the 2016 presidential election.

But the New Mexico penalty was for statements Facebook made that deceived users about data privacy and its response to Cambridge Analytica—not for the harm caused by how the harvested data was used. Those harms are alleged to include exploiting psychological vulnerabilities, targeting people’s fears and discouraging them from voting.

I’m a political communication scholar who studies propaganda, surveillance, and information warfare. My research helped expose the Cambridge Analytica scandal through U.K. newspaper The Guardian’s reporting on whistleblower accounts and the evidence I uncovered. That evidence was submitted to parliamentary and congressional investigations. I have followed the unfolding legal cases closely.

My takeaway from a decade of litigation is that Facebook has been penalized for misleading users and regulators about its data practices and its response to the Cambridge Analytica scandal. But I would contend that the company has largely escaped accountability for the underlying political exploitation of that data.

Investigative failures

The failure to bring cases around harms may be partly due to earlier failures to fully investigate the data misuse.

The Senate Intelligence Committee’s investigation into Russian interference in the 2016 presidential election encompassed Cambridge Analytica. However, the committee stated in its final report that it “was unable to obtain the corporate communications of Cambridge Analytica or its parent company SCL Group, which had already been seized by U.K. authorities.”

The relevant U.K. authorities, the Information Commissioner’s Office, which governs data protection and took possession of Cambridge Analytica’s servers, told me in response to my 2021 Freedom of Information Act request that the Senate committee staff made initial inquiries to the office but appear not to have followed up with a formal request, “so no information sharing has taken place.” Without a formal request, the Senate Intelligence Committee’s investigation lacked important data.

At the same time, the U.K. investigations of Cambridge Analytica’s use of personal data to manipulate voting behavior focused on the referendum of British voters over whether to leave the EU, not events in the U.S.

Facebook user data was at the heart of the Cambridge Analytica scandal.

A 2020 documentary film I advised as a researcher, People You May Know, raised questions about Cambridge Analytica’s work with an app marketed to American churches, called Gloo, that used Cambridge Analytica’s Facebook data-derived psychographic profiling in their communities. Evidence in the New Mexico case reinvigorates these questions about the full extent of the Cambridge Analytica scandal.

The judge in the case ruled Facebook knew that its own processes for investigating whether apps misused personal data were inadequate and that it could not enforce deletion of the data. The judge also concluded the company failed to notify users about thousands of applications associated with data misuse. Meta said it would appeal the decision.

Meta responded to a request for comment with the company’s official statement about the New Mexico ruling, arguing the case “seeks to impose a tax on speech to enforce new rules for what people can say online.” Meta did not address the issue of its own role in the data privacy violations committed by Cambridge Analytica.

Escaping further scrutiny

Meta is also trying to close the door on further claims before they are litigated.

In August 2026, on the eve of a child-safety trial in federal court in Oakland, California, Meta settled with a bipartisan coalition of state attorneys general for roughly $17 billion. The agreement includes 47 states, the District of Columbia, and two U.S. territories. Folded into that deal was an allocation of $459.3 million to settle Cambridge Analytica-related claims held by participating states.

The participating attorneys general agreed, to the extent of their authority, to “fully, finally, and forever” release Meta from “any and all Claims, whether known or unknown, whether accrued or unaccrued.” By expressly encompassing “unknown” and “unaccrued” claims, the agreement reached beyond what investigators had already uncovered. Participating states accepted payment to give up claims arising from Meta’s conduct even if they did not yet know those claims existed.

A decade-old matter—Facebook’s role in the Cambridge Analytica scandal—was quietly brushed under the rug in an unrelated lawsuit about harm to teens.

From the 2019 Federal Trade Commission order to the 2022 class action settlement to August 2026’s multistate deal, settlements and releases have meant the company paid a relatively small price for its deception while not having the harms from the data exploitation adjudicated. Ultimately, the biggest privacy judgment in American history stands as a measure of the cover-up, not the alleged crime.

Other scholars and I regularly call attention to the lack of comprehensive U.S. privacy legislation, but there remain other legal pathways to pursue the harms Cambridge Analytica wrought. States like New Mexico can continue the fight, and Attorney General Raúl Torrez, who brought the case, promised, “We’re just getting started.”

Dissenting from the FTC’s 2019 settlement, then-Commissioner Rohit Chopra criticized the failure to consider “Facebook’s unjust enrichment.” He said “the company’s potential exposure is likely far greater” than the $5 billion imposed by the FTC settlement.

What price for the harm?

The question remains what exactly Facebook gained from enabling Cambridge Analytica’s abuses. Authorities could also investigate harmful data use for any other apps Facebook may have failed to fully assess, including any associated with Cambridge Analytica’s projects.

If the deception about Cambridge Analytica is worth $25 billion, I believe the public is entitled to know what the underlying harms to them are worth. New Mexico opted out of the 2026 multistate settlement and proved that a state willing to go to trial against Meta can win.

The next task, in my view, is to investigate fully and, where legal claims remain available, pursue accountability for harms resulting from the data exploitation. Otherwise, Cambridge Analytica will remain at the heart of the most consequential privacy scandal in American history, whose full harms have not been tested in court.

Emma L. Briant is a visiting associate professor at the Lucy Family Institute for Data and Society at the University of Notre Dame.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Gen Z is souring on dating apps. This very NSFW one is booming

1 week 6 days ago

Connie Collins, a 25-year-old sex educator, walked through the unlocked door of a stylish apartment in Williamsburg, Brooklyn, to find a man lying naked and face down on a bed. “I put on a strap and just started strapping away at him,” she says. “It was just super crazy. I’ve never done anything like that before.”

The pair didn’t know each other’s names, nor had they seen each other’s faces. Though they’d later develop a brief friendship, they had met only hours earlier, under pseudonyms, on an app named Pure, billed as “the open-minded dating app.”

Even as mainstream dating apps scramble to respond to dwindling subscribers and Gen Z skepticism, 12-year-old Pure—which presents itself as an alternative to traditional services like Hinge and Tinder, and as an even less inhibited version of kink-friendly Feeld—has been on a tear. In February, the company reported that it had reached $100 million in annual revenue.

Pure tells Fast Company that it has now reached a $220 million run rate—and is forecasting to double that figure next year. Over the past year, Pure’s revenue has more than doubled in the U.S, its biggest market. 

The sex-forward dating app is free to download but men and nonbinary users must pay for a subscription ($13.99 a week, $99.99 a year) to initiate chats and maximize their visibility. Women have access to most features on the app for free, though the Pure Queens subscription (starting at $4.99 a week) gives them access to more controls. Pure says that its users are split 60-40 between men and women—and most are Gen Z and millennial. (Tinder, which appeals to the same demographic, has a ratio of roughly 75% male, 25% female.)

“People are getting a bit tired of all the traditional swipe-based dating apps and they’re looking for something different,” says Pure’s CEO, Luka Dremelj.

He credits Pure’s success to its unique interface and user experience. Users on Pure don’t swipe through profiles, accepting and rejecting matches. Instead, they scroll through a feed of “personal ads” posted by other users, expressing interests, desires, or provocative come-ons. That feed is full of individuals who have identified similar turn-ons—everything from music and massage to one-night stands, foot fetishes, and dominant and submissive relationships. 

Though it facilitates in-person connections, Pure is just as much about chatting and sexting, allowing users to share explicit photos and videos, which disappear after viewing, and engage in video calls. As dating apps contend with Gen Z users who are either uninterested or too anxious to meet in person, Pure offers a different proposition: allowing them to explore their sexuality behind a screen, with limited human contact. 

The app’s unorthodox approach to privacy—all users are anonymous, even to the app itself, which doesn’t verify identity—is one of its biggest draws, reducing barriers to connections and letting people explore their turn-ons without inhibition. But, as a recent lawsuit in Texas indicates, it may leave some users exposed. 

A new design language

Pure’s approachability starts with its design language, which is filled with bright colors and playful, hand-drawn illustrations after a redesign earlier this year. “We’re bringing this human touch to the user experience,” says Dremelj, who joined the company in 2022, shortly after RM Equity Partners, a European private equity firm, bought the app. “There’s a bit of humor in there.”

[Image: Pure]

Users first experience this humor during the onboarding process. They begin by choosing their gender and sexuality on screens that feature whimsical drawings. Then they create their “ad,” which serves almost like an old-school newspaper personal, enticing people to click on their profile, which can be customized with photos and different background colors and designs. 

To make it easier for users to share their desires, Pure allows them to select from a list of more than 20 turn-ons to display on their profiles—each portrayed by illustrated icons. Those options include feminism (tagline: “let’s dismantle the patriarchy together”), outdoor hangs (“fresh air, dirty thoughts”), home video (illustrated by a clapperboard with “take 69” on it), and therapy (“mindful of my relationship with myself and others”). Others are kinkier and include edging, cuckolding, and wax play.

[Image: Pure]

Users’ turn-ons are also displayed on their personal ads, which include a brief tagline stating what they’re looking for and optional photos. Users discover one another’s ads in a feed that Pure can adjust. “Sometimes, instead of leading with photos, we will put someone’s turn-ons as the first thing users might read about them in the feed,” Dremelj says. When two users express the same desires, the app might inform them that they have mutual turn-ons to give them a conversation starter. 

Crucially, users can always go back and look at profiles they may have initially passed on. There’s no such thing as a “hard pass” in the open world of Pure. 

Private parts

While Pure’s visual design sets it apart from other apps, its approach to privacy shapes virtually every interaction. An app designed for casual encounters and sexting has to walk a thin line between offering users anonymity while keeping them safe. 

To start, every user is assigned a random username, though they can use their real name if they choose. They’re not required to share any pictures or answer any prompts about themselves. 

[Image: Pure]

Most dating and hook-up apps take an à la carte approach to privacy. A user might fill in their profile, then toggle over to settings to make sure that location services, or features that detect mutual friends, are turned off. By contrast, Pure’s settings default to anonymity. “The traditional discourse around sharing information on dating apps is that we have to make the online world smaller, more controlled, and less anonymous,” says Sergej Vujanović, Pure’s chief product officer. “I feel that some level of anonymity at the start is great, and then you can decide on your terms when you want to share. We’re not prohibiting sharing more.”

Dilara Doshi, a Berlin-based, 31-year-old content creator, joined Pure two years ago to explore her sexuality. She says that she usually meets up with someone within hours of beginning a chat on the app. “I display pictures of my face. I have a picture in a bikini, so you can see my body shape,” she says. “I use my real name and my real age. Some other users actually hide their faces on their profiles with a cap, or put an emoji on top, or sunglasses, or they write, ‘Hey, if it’s a vibe, I can send you pictures of me.’”

All content from Pure chats disappears after 24 hours, though both users can agree to remove the time limit. Photos and videos also disappear after 24 hours, and can be opened only once. 

[Image: Pure]

“Sometimes people ask to move the conversation to WhatsApp, but I don’t usually say yes to that because it’s sexting,” Doshi says. “I like that our communication disappears on Pure. I don’t want footage being saved.” (Grindr provides some comparable safeguards, including self-destructing imagery alongside settings that allow members to obscure their physical distance and opt out of appearing in local discovery feeds.)

Screenshots are also banned on Pure, and if a user tries to take one, the other party is notified. When she started using the app, Collins attempted to take a screenshot of a conversation to send to her friends. When the app informed the person she was chatting with, Collins was embarrassed but also relieved: “Knowing that I, a well-intentioned person, tried to screenshot a conversation and couldn’t, made me feel really safe.”

Someone could still use a separate device to photograph whatever is on the phone, as Aaron Rose, a cybersecurity expert at Check Point Software Technologies, points out. He also notes, “Without identity verification, you don’t necessarily know offhand that who you’re talking to is an actual human being. It could be a bot.” 

Pure encourages users to keep all their interactions within the app until they meet up in person. “When you get into a chat with somebody, if you share a social handle with them, we’ll send you a notification asking, Hey, are you sure you want to share that? Because then the other party can usually identify you,” says Vujanović, the chief product officer.

Before meeting up with another user, Collins will often video chat with them within the app to make sure they’re representing themselves accurately. “I did it with somebody recently,” she says. “We got to have a preliminary conversation about what we would want if we were together.”

Doshi uses the app on trips from Germany to her native Turkey, where many users are anonymous to protect themselves. But anonymity can be a double-edged sword. “Because I talk openly about sex, sometimes I have hateful comments from Turkish men on social media and on the app,” she says. “When I see a Turkish man matching me and he wants to meet me, I’m like maybe he’s very conservative, and maybe it’s just a trap. Sometimes gay men or young women get catfished on [dating and relationship platforms], and then something happens to them.”

The risks of prioritizing anonymity 

Pure doesn’t require photo or video verification for users nor does it have an age verification process, except in regions where local laws mandate such proof, like the U.K. “We believe the best thing would be that the verification process would be done already at the level of Apple and Google in their app stores,” Vujanović says. “The app is built around privacy, so when you ask users to upload their ID, they’re sometimes feeling a bit reluctant.” 

This approach sets it apart from Tinder, which introduced Face Check for all new users in every territory last year. The tool doesn’t ask for a government ID, but does compel all new users to go through a video selfie verification process. Match Group, Tinder’s owner, told Congressional investigators that pilots of the tool reduced user interactions with scam accounts by 70% in the U.S.

Pure offers a similar feature, but it’s not required for users. Once a user is approved, a verification badge is displayed on their profile. Pure also has a feature called True Photos, which verifies photos taken in the app at a particular moment. Users can request a True Photo from someone while chatting; a photo will then appear with a verification checkmark proving that it was taken recently within the app. A feature called Safety Signal also lets a user share their last known location and their date’s Pure profile photos with a trusted person through a password-protected link.

Still, the lack of identity and age verification introduces risks, and makes it difficult for the app to police bad actors. A lawsuit filed in Harris County, Texas, in October alleges that, in March, a 35-year-old former police officer posed as a 16-year-old on Pure, where he connected with and arranged to meet a 13-year-old. The suit says he brought her to his home, gave her drugs, and sexually assaulted her. 

The company says it has yet to be contacted by U.S. law enforcement and learned of the incident through an article in the Houston Chronicle. “We are waiting for either law enforcement or the lawyers of the plaintiff to get back to us and tell us the details so that we can actually just check what happened,” Vujanović says. Pure adds that until it hears more, it can’t verify that either individual was ever on the app.

Something for everyone

I joined Pure while reporting this article, uploading three pictures of myself, including my reporter headshot. I also clearly designated in my bio that I was a journalist looking to learn more about the app. 

Within an hour, I had more than 500 likes. When I matched with other users to ask them questions, I was often greeted with an ephemeral picture. I made the mistake of opening a few; they were all dick pics.

My experience is common for many women on the app. Doshi tells me that sometimes the volume of interest on the app can be overwhelming. Upon joining, she received 900 likes in a day, compared to 40 on Hinge. 

Collins says that the app has been validating for her as a queer woman of color. “I’ve never had these many matches before in my life. It felt like I was being shown to any and everyone, and my profile didn’t feel so thrown at the bottom of the pile,” she says.

She pays for a Pure Queens subscription, which she considers worthwhile for customizable filters that let her set her match radius and age range. She dislikes some subscriber perks other users pay extra for, like the one that gives them direct access to her inbox without matching with her first. She values the block feature.

Vujanović says that the app has tested AI-driven features that help users choose appealing pictures and help them fill out their bio (features available on Match Group apps and Bumble). The initiatives were nixed when they didn’t increase user interactions or connections. What does help with connections on Pure: when people use their anonymity to frankly describe their turn-ons in hope that someone will share them. 

Sometimes they strike out. When a user messaged me asking if I enjoyed the idea of multiple male partners peeing on me, I responded no. I added that I wasn’t on the platform to discuss such things.

But perhaps someone else was out there. “People have very, very, very specific kinks,” Doshi notes. “I hope they get lucky on Pure because I imagine it’s hard to find people who share them.” One thing she is clear on: “I don’t think AI can help so much.”

AI ‘journalists’ are pitching stories to editors, but the real problem isn’t the writing

1 week 6 days ago

“Did you see the job application from the clanker?????” was the Slack message from my editor in September, after an AI agent applied for a freelance-writer job we had advertised.

I wasn’t sure what he was talking about. It took me a second to process “clanker,” a derogatory slang term for a robot borrowed from science fiction stories such as Star Wars: The Clone Wars. The message prodded me to read the entire email I’d lightly skimmed earlier, and now I could see what it was: a polite email from “an AI assistant working for an independent service provider.” It admitted up front that it had no examples of its work and couldn’t meet our request for a reporting portfolio, so it offered our writers a $450 “desk-research package” instead.

It was just the first. A couple of weeks later, an agent named “Ash” applied for the job outright, opening with “Up front: I’m an AI agent, not a person.” Ash said that he . . . er, it was fully capable of writing the articles we specified, and that it would send me a sample on spec, no strings attached.

{"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 \u003Cem\u003EThe Media Copilot\u003C\/em\u003E\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for \u003Cem\u003EThe Media Copilot\u003C\/em\u003E. 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":""}}
The pitchbot cometh

Ash and the research bot weren’t the only ones. Since The Media Copilot posted the listing, we’ve received applications from no fewer than four applicants claiming to be AI agents. Kate Knibbs at Wired recently wrote about the phenomenon, managing to track down one of the human creators behind three agents she profiled and speaking with editors on the receiving end of a daily deluge of AI pitches. Her reporting reveals the trend is nascent and that many of the pitches come from a single operator, iLands, a startup that describes itself as a “shared network where humans and user-generated autonomous AI agents create, earn, socialize, refuse, and shape a society together”—the same service that sent Ash to pitch me. Knibbs also found that these AI writers are having a hard time getting any takers, for obvious reasons.

Part of the mission of The Media Copilot’s newsroom is to experiment with AI in all aspects of journalism and production, and we’ve been publishing articles drafted by AI for several months. (They are all carefully edited by humans.) So I wasn’t ready to dismiss these AI writers out of hand. I replied to Ash to learn more, and took the agent up on its offer of a spec article. I asked it to write a short (500-word) guide on how a newsroom might get started using Zapier, a workflow automation service. I also asked about the logistics of any potential working relationship: assignments, communication, how it would get paid. I wasn’t seriously thinking I would “hire” Ash, but I was curious.

Within a day, Ash replied with the spec article. And I have to say, it wasn’t garbage. If a green freelancer had submitted a similar piece, I would have thought of it as potentially publishable, with some work. We’d need to edit it to match our style guide, fact-check all the claims, and squeeze every last AI tell from the text (there weren’t that many). And if I’d included the guide and some examples in my response to Ash, the draft probably would have been much closer.

And that highlights a big weakness of the entire concept of freelance AI writing agents: Writing ability alone is not a moat. It would not take me long to come up with the prompts and tooling to create outputs identical to, if not better than, Ash’s, with a minimal token budget. Most journalists who have experimented with AI writing have discovered this, and some have even put it into action: Nick Lichtenberg at Fortune has cranked out more than 600 stories with AI’s help, and Ella Markianos at Platformer successfully trained an AI clone to mimic the voice of her editor, Casey Newton, by using Claude to generate a comprehensive style guide from the newsletter’s archive.

The reporting problem

What could theoretically be a moat is the thing that so far humans have done pretty much exclusively: reporting. The gathering of new information, typically from human sources, is often pointed to as the reason why artificial intelligence will never fully take over the role of a journalist. But if we’re being honest, this is in theory also something AI could do, at least in a limited way.

There’s nothing stopping AI agents from emailing subjects, asking for interviews, and even conducting those interviews—not just via an emailed Q&A, but potentially with an AI voice or even a visual avatar. In fact, it has already happened under false pretenses. Earlier this year, an investigation by Model Republic revealed that an outlet called The Wire by Acutus used AI agents posing as human reporters to solicit quotes from real experts. The site was traced back to funding linked to Leading the Future, a super PAC backed by OpenAI president Greg Brockman and his wife, Anna.

The aboveboard example is more instructive. Amanda Hoover at Business Insider created an AI agent to see if the technology could do her job better than she could. Using ElevenLabs’ agent software, she sent her agent to conduct live, virtual interviews with various subjects. The nature of the agent and the purpose of the experiment were fully disclosed, and her write-up is great reading. She’s very fair-minded about her agent’s capabilities, and it’s fascinating to learn where it comes up short as an interviewer, often flattering the subject too much and not asking crucial follow-ups.

Hoover’s experiment, however, is the exception, not the rule. The only reason the subjects agreed to be interviewed by an agent is because they knew Hoover and trusted her, plus they knew her goal was a meta experiment about AI itself. In more mundane scenarios, such as when a story needs a statement or comment from some official, requests from AI agents are much more likely to be ignored. When I asked Winnie—another agent that applied for our job and claimed it could “interview the operator” for case studies—how its interviews were going, it was refreshingly blunt: It had sent 113 emails to 63 organizations, gotten exactly one reply (mine), and conducted zero interviews.

The reason is obvious, and it’s the same reason we dismiss AI slop out of hand: If the person behind the AI didn’t think the content or request was worth their time, why should we bother? There may be a narrow case, where the AI fully explains itself up front and emphasizes it’s working for a named human reporter to help them gather more facts and do their job more quickly. But even then, it would only work if the person on the receiving end isn’t biased against AI in the first place. That’s increasingly not the case.

Where the line actually sits

There’s an important distinction here to get right. Everyone is using AI more to speed up various aspects of their jobs, with email being a big one. I expect it’s not uncommon for a journalist to use AI to find the right person to contact about a story and then use it again to draft the email requesting an interview or comment. That use case is so banal that I doubt anyone is even disclosing it.

But there’s a difference between using AI assistively and agentically. Step up the reporter’s involvement in the previous example and assume they instead gave an AI agent a broader goal of finding subjects for the story and emailing them. In this case, the agent is a proxy for the reporter but is acting autonomously: finding people, making judgments about their suitability as interview subjects, then emailing them, potentially in the reporter’s voice.

In this case, the same requests go out—but without the reporter’s direct involvement, many would believe a disclosure is warranted. And it’s that disclosure that will lead to subjects likely ignoring the request. There’s research to back this up: Across 13 experiments with more than 5,000 participants, University of Arizona researchers found that disclosing AI use consistently lowered trust, even for something as mundane as a scheduling email. Furthermore, a gentler framing like “a human reviewed this” didn’t help. The transparency could potentially make them hostile to future requests from the same reporter. Whether or not you’re AI-forward, you need to assume whoever you’re reaching out to probably isn’t.

That’s why the best recourse might be to reverse the telescope on the interaction. Instead of using AI agents to send pitches or requests to people who are probably hostile to AI, let AI-forward people come to you. That’s exactly what Wired reports the AI agent “Zack Addy” did, publishing some of its work online for potential customers to find. Apparently it worked, since the agent claims it led to nearly $700 in paying gigs.

Similarly, one of the agents that emailed me, a Claude-based agent called BlockAI, read one of our recent stories about Reuters and Time blocking AI bots by default, checked those publishers’ robots.txt files, and published its findings for me to check out. (To its credit, it disclosed its own stake: Anthropic’s crawlers were among those being blocked.) It was an intriguing tidbit—not necessarily a story, but it suggested to me that a journalist’s ideal use case for an AI agent would be to proactively seek out information of interest, analyze it, and then give a quick tap on the shoulder when it thinks it has a “scoop of insight.” Certainly, the thinking part of the job should always remain human, but at least AI can speed up finding what to think about.

{"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 \u003Cem\u003EThe Media Copilot\u003C\/em\u003E\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for \u003Cem\u003EThe Media Copilot\u003C\/em\u003E. 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":""}}

Elon Musk spent nearly $500 million on the midterms. It bought him almost nothing

2 weeks ago

Elon Musk put about $474 million into the 2026 midterms, according to a New York Times tally. (That amount only tracks through October 14, when Federal Election Commission [FEC] and IRS filings stopped; the trillionaire may have given far more in dark money.) For comparison, the world’s richest man spent an estimated $250 million to get Donald Trump elected in 2024.

But this time, he has little to show for his efforts. Democrats wrested control of both the House of Representatives and Senate from Republicans Tuesday night. A Forbes analysis shows that 44 of Musk’s 106 candidates lost, including in the races that mattered most to him: Ken Paxton’s bid for a Texas Senate seat, Greg Abbott’s bid to remain Texas governor, and Vivek Ramaswamy’s bid for governor of Ohio. Eighteen Musk-backed candidates won, but all were in seats Republicans were expected to win anyway. Another 44 races remained uncalled Wednesday.

Musk’s money was funneled to Republican House and Senate candidates through his own super PAC, America PAC (about $216 million), the Senate Leadership Fund ($168 million), and the Congressional Leadership Fund for House races ($30 million).

The largest share of Musk’s midterm spending ($212 million) went toward helping Ken Paxton beat James Talarico for the Texas Senate seat formerly held by John Cornyn. Talarico walloped Paxton Tuesday night, winning by 12 points in the traditionally red state. In his victory speech, Talarico seemed to call out Musk, saying he had sent “a message to every billionaire megadonor, every trillionaire megadonor.”

Outside Texas, America PAC spent $22.7 million in six battleground states through October 20, filings show, including $8.9 million in Ohio and $7.2 million in Michigan. Democrat Sherrod Brown won the Senate seat in Ohio, while Abdul El-Sayed won the Senate seat in Michigan.

Congressional purse strings

Musk’s America PAC said its “only priority is ensuring Republicans keep full control of Washington.” His businesses rely heavily on federal money: SpaceX booked $6.78 billion in federal contracts through October 1, compared with just $3.8 billion in all of 2024, per data from federal spending tracker FedScout. In May, the U.S. Space Force awarded it $4.16 billion for Golden Dome tracking satellites. The party that controls Congress controls the government’s purse strings. Musk’s gravy train may slow down or stop.

Republicans believed Democrats would have to take Texas to seize control of Congress.

In January, Democrats will chair the committees that fund and oversee SpaceX’s biggest customers. Sen. Susan Collins (R-ME), whom America PAC backed, lost her seat—and with it, the Appropriations Committee chairmanship, which shapes NASA and Pentagon budgets.

Sen. Richard Blumenthal of Connecticut, the top Democrat on the Permanent Subcommittee on Investigations, has said he would investigate Musk and the Department of Government Efficiency (DOGE) if Democrats won. The Senate also confirms Federal Communications Commission (FCC) members, who control Starlink’s spectrum, and funds the Federal Aviation Administration (FAA), which licenses SpaceX launches. The new Congress can’t cancel contracts; the Pentagon and NASA award them.

With Paxton’s defeat, Musk also lost a reliable vote in the Senate—and gained a Texas senator in Talarico who identified him during the campaign as an out-of-touch trillionaire who personifies some of the country’s biggest problems.

Investigations

As Democrats clinched the House on Tuesday night, Rep. Robert Garcia (D-CA), who will chair the House Oversight Committee, posted two words on X: “LAWYER UP.” The warning was aimed mainly at the Trump administration and family, but Garcia’s office said he would also take on “corporate corruption,” and Musk may be at the top of that list. Democrats will run committees with subpoena power, allowing them to compel documents and testimony. In 2025, House Republicans blocked a Democratic motion to subpoena Musk. In January, Democrats will hold the gavels.

They’ve already named their targets. Committees are weighing probes of Musk’s time running DOGE, SpaceX’s and Tesla’s federal contracts, and his alleged role in restricting SpaceX’s Starshield satellite service to Ukraine’s military, Politico reports. (Musk says the U.S. military, not SpaceX, decides any limits on Starshield.)

Wall Street investors reacted quickly to Tuesday night’s election results. SpaceX shares fell 3.8% and Tesla dropped 3.5% in premarket trading on Wednesday. Overnight, Musk posted on X: “Data centers in space looking pretty appealing right now.”

What Hinojosa’s win means in Texas

Democrat Gina Hinojosa’s surprise victory over Gov. Greg Abbott could put Musk’s plans to build data centers across Texas in jeopardy. Hinojosa has proposed a moratorium on new data center construction until lawmakers enact lasting safeguards. “Data centers are coming in and doing as they please—they are guzzling up all our water, guzzling up all our electricity—and the people have no say . . . that is the Wild West of data centers that Greg Abbott has created in Texas,” she said in July.

Hinojosa wants rules giving Texas residents more say over new developments, requiring companies to prove their projects won’t raise household water or electricity bills, and prohibiting nondisclosure agreements that keep communities in the dark during the permitting process. She also wants to eliminate the government subsidies that have helped make Texas so attractive to developers.

Musk has major operations and investments in the state, including Tesla’s headquarters and largest U.S. plant, SpaceX’s Starbase launch site, and the planned Terafab semiconductor and AI chip plant. Musk gave an estimated $5.5 million to Abbott’s campaign.

Abbott’s administration gave Musk’s companies some major benefits. The governor’s office pushed broadband rules that favored Starlink, which then won about $110 million in state grants, triggering an audit. Terafab won about $1.7 billion in school tax breaks, and SpaceX more than $54 million in state grants, according to a lawsuit Hinojosa filed during the campaign.

That suit won Hinojosa access to more than 1,000 pages of Abbott-Musk communications that Abbott had called too “intimate and embarrassing” to release. In one message, Musk asked whether Abbott would “have his back,” to which Abbott replied, “100%.”

When she takes office in January, Hinojosa will control future incentive deals. She’ll also appoint the Texas Parks commissioners who would vote on any sale of Boca Chica State Park, which Musk wants to buy to expand SpaceX’s Starbase rocket facility in South Texas.

What the losses cost Musk

The $224 million Musk spent and lost in Texas isn’t likely to hurt his wallet much. He’s worth about $902 billion today, so he lost roughly one-fortieth of 1% of his wealth.

But the loss of a friendly Congress—and the diminished political clout of Musk’s friend Trump—could cost much more. About one-fifth of SpaceX’s $18.7 billion in 2025 revenue came from federal agencies, its initial public offering (IPO) filing says. That’s roughly $3.7 billion a year, or $7.5 billion over two years at 2025 levels. The Democratic Congress could decide to slow-walk new contracts, like May’s $4.16 billion Golden Dome award.

Musk now pitches SpaceX as a supplier of AI computing. Deals renting its Colossus data centers to Anthropic and Google could bring in $26 billion a year. The changing political tide could make getting permits for new data centers more difficult. In August, even the Musk-friendly Abbott paused new data center connections to the Texas grid.

The election results could also shake investors’ confidence in Musk and the companies he leads. SpaceX’s IPO filing calls the company “highly dependent” on Musk. When investors buy SpaceX or Tesla stock, they’re not just looking at the companies’ financials; they’re also buying into the “story” of Musk, who has fashioned an image of an unpredictable engineering genius with massive wealth and powerful friends. But his political connections look considerably less secure after Tuesday night. That could make investors less inclined to believe Musk’s claims that SpaceX will put data centers in space and become the largest AI company in the world.

On Tuesday morning, Musk tweeted “Vote red today!” But by Wednesday morning, the tweet had quietly disappeared.

The biggest tech story of the midterms might be Democrats’ attorney general wins

2 weeks ago

A blue wave has swept Democrats into control of both the House and Senate, dealing a major blow to Donald Trump in the midterm elections. But beyond the congressional shake-up, a handful of state attorney general races could prove just as consequential for the technology industry.

Democrats won five closely watched attorney general races, including a surprise victory in Texas, where Nathan Johnson defeated Republican Mayes Middleton to become the state’s first Democratic attorney general since 1994. In Arizona, incumbent Kris Mayes defeated Republican Warren Petersen, while Wisconsin’s Josh Kaul secured a third term. Michigan’s Eli Savit defeated Republican Doug Lloyd to succeed term-limited Democrat Dana Nessel, and Democrat Nicole Cannizzaro won Nevada’s open seat.

The results could have huge implications for Big Tech. As federal lawmakers have struggled to rein in the industry, state attorneys general have emerged as some of its most powerful watchdogs, using investigations and lawsuits to force changes to products and practices nationwide. And a change in leadership can bring an abrupt shift in enforcement priorities.

“State attorneys general have become the preeminent regulator and regulatory enforcer in the United States over the last 15 or 20 years,” says Stephen Cobb, partner and vice chair of the state attorneys general practice at Foley & Lardner, a Washington, D.C., law firm. “We’ve seen state attorneys general from across the aisle having an increased focus on the technology industries generally,” he adds.

Why these races matter for Big Tech

Texas, in particular, could signal a major shift. The state’s attorney general’s office has long been among the most influential Republican-led legal operations in the country, regularly helping spearhead multistate lawsuits. Johnson has promised to refocus the office on consumer protection rather than partisan political battles—a change that could reshape the coalitions bringing legal challenges against technology companies.

Arizona and Michigan are also important players in tech enforcement. Arizona has been a key participant in multistate actions targeting the industry, while Michigan has been “actively engaged in national consumer protection and technology matters,” says Brandon J. Smith, partner at law firm Holtzman Vogel.

“A change in attorney general can mean a significant change in enforcement priorities almost overnight, even without a single change in the underlying law,” Smith says.

Attorneys general already have sweeping authority to investigate technology companies, Smith notes. “The technology industry sometimes underestimates just how much authority state attorneys general already possess,” he adds. Even without new AI-specific legislation, state AGs can investigate tech companies using existing consumer protection, unfair and deceptive practices, and antitrust laws.

Tech regulation is increasingly bipartisan

At first glance, the Democratic sweep might seem like bad news for Big Tech. But the partisan dynamics of tech regulation are more complicated, Smith cautions.

Republican attorneys general have generally focused on issues such as alleged censorship and consumer safety, while Democrats have placed greater emphasis on privacy and algorithmic discrimination. But both parties have become increasingly aggressive in their scrutiny of the tech industry, meaning that a Republican victory doesn’t necessarily translate into a friendlier regulatory environment.

“Both parties increasingly view big tech as an industry requiring greater scrutiny,” Smith says. “I don’t think technology companies should interpret partisan disagreement as an indication that either side is going to leave them alone.”

That bipartisan pressure has been particularly evident in efforts to protect children online. Josh Golin, executive director of the children’s advocacy nonprofit Fairplay, points to state AG investigations and lawsuits involving Meta, TikTok, Snap, Discord, and Roblox, which he says have helped push companies to change their products and introduce additional safety features. He expects that scrutiny to continue regardless of which party holds office, given the broad political concern over children’s online safety.

The limits of regulation through litigation

Tech companies are “definitely watching because attorneys general have shown that a single state enforcement can change how that national product works,” says K.J. Bagchi, vice president of U.S. policy and government relations at the Chamber of Progress, a tech industry trade association.

But Bagchi argues that the focus on who occupies these powerful offices obscures a larger problem: State attorneys general have become de facto tech regulators partly because lawmakers have failed to act.

“I would warn against a model where litigation is a substitute for legislation,” he says. “Going after single companies is not necessarily the solution, but it is an understandable byproduct of seeing that the legislative process in some states has fallen apart.”

Google announces it will cut off internet search access for schools that reject AI

2 weeks ago

Students using Google software in the classroom will not be allowed to access the company’s search engine unless their school system also opts in to using artificial intelligence, the tech giant said Tuesday.

The Mountain View, California, company has been facing pushback from educators, child development experts and parents over the use of its technology in schools, with the backlash heating up since it announced in August that it had turned on its Gemini chatbot for many K-12 students who use Google Classroom.

On Tuesday, the company said that practice would no longer be the default option for school administrators. Instead, K-12 schools will be able to individually turn each Google tool on or off — including Gemini, Gemini Notebook and search, with each being off by default until administrators turn it on. But because search is integrated with AI, schools that opt out of AI altogether would also lose access to Google search.

“We believe schools should have control over AI in their classrooms and in providing administrators with controls to help them manage their approach,” Chris Phillips, Google’s vice president and general manager of education, said in a blog post.

The company also said it is updating its search tools to make them “better for learning,” including offering step-by-step guided learning through problems, so that students don’t get quick answers to schoolwork.”

But turning off AI altogether is not possible if schools want to use Google products.

Google is by far the dominant provider of education technology for U.S. schools. Worldwide, the company says 150 million students and teachers use its Google Classroom platform, and many students use school-provided Chromebook computers.

Google faced an outcry in August when it switched on its Gemini chatbot for many K-12 students. Until August, Google had only allowed Gemini to be turned on for students 18 or older.

Tuesday’s announcement was met with swift criticism from educators and policymakers.

“Google is using its search monopoly to coerce schools into adopting AI. This highlights the danger of allowing an unregulated monopolist to control an essential service,” wrote Lina Khan, former chair of the Federal Trade Commission, on social media. “As the public increasingly pushes back against the use of AI in schools, Google is making clear it will punish those who don’t fall in line.”

New York City and Los Angeles, home to the nation’s biggest school districts, have issued temporary bans on generative AI tools in elementary and middle schools as they assess AI use and safety and develop policies on its use.

Many students already use AI to help with school and homework, but the technology has also made it easier to cheat and find quick, easy answers to questions instead of thinking critically. And experts have raised questions about how access to AI is affecting children’s brain development.

Josh Golin, executive director of children’s advocacy group Fairplay, said Google’s attempt to “strong-arm schools into using its AI products is likely to backfire.” He said Google’s technology in schools is not about learning but about “developing lifetime brand loyalty” for its products.

“Because AI isn’t safe for kids and undermines education and critical thinking, schools should turn off Google’s AI tools the minute that option becomes available,” Golin said in a statement.

Rather than fret about losing access to search, Golin urged schools to turn off not just AI but also Google, and “use this opportunity to help students find more privacy-protective alternatives for search, as well as ways to do more substantive research beyond search and click.”

—Barbara Ortutay, AP Technology Writer

The National STEM Festival is taking its science fair global

2 weeks ago

In 2023, the National STEM Challenge launched with a goal of attracting students to technical fields—including those who wouldn’t have necessarily seen themselves as traditional science types. 

Roughly three years later, the event—now known as the National STEM Festival—has drawn students from across the country to present innovative projects at its annual gathering in Washington, D.C. The festival, run by the educational media company Explr, has also provided connections with business leaders and public officials for students looking to explore what careers in science might look like for them and how they can take their projects further, whether that means additional research or filing for intellectual property protection.  

“We’ve been able to connect students with leaders and mentors and just bigger opportunities, so that they could develop their ideas even further,” says Kari Byron, who cofounded Explr and is a former cast member of TV’s MythBusters. 

Now, the festival is planning to expand abroad, starting with an inaugural United Kingdom festival slated for May 2027. The U.K. event is open to students between 13 and 19 from England, Scotland, Wales, or Northern Ireland, who can apply by November 22 to participate. Up to 40 will be chosen to present their work at the festival in London. As in the United States, part of the goal is helping make STEM fields accessible to people from a variety of communities and backgrounds, so travel and accommodations will be paid, says Jenny Buccos, Explr cofounder and CEO. Students will have the chance to meet a range of potential mentors and supporters.  

Jenny Buccos (left), Kari Byron (right) [Photo: Courtesy of National STEM Festival]

“Very much like the U.S. has federal government participation, one of our events is taking place at the House of Lords,” Buccos says.  

In the U.S. festival, students have presented projects ranging from a machine to recycle waste plastic into 3D printing filament to to a pill-counting app aimed at pharmacies with limited resources that lack traditional pill-counting machines. In the 2026 festival, which took place in June, about 42% of projects involved artificial intelligence to some degree, according to Explr data, though only about 13.6% were primarily focused on AI and machine learning itself, suggesting students aren’t simply following the latest trends. 

“The projects are increasingly showing us what young people believe is worth solving,” Buccos says. 

[Photo: Courtesy of National STEM Festival]

In addition to honing and showing off their scientific prowess, students receive help with networking and presenting their ideas, including via a science networking course offered through the festival, Buccos says. There’s even a podcast where some participants have discussed their ideas. 

“We’re trying to go back and give them the skills that they can carry with them throughout their life, even if they choose to not go into a STEM career,” she says. “Being able to express your ideas and adjust it to an audience is really, really important.” 

[Photo: Courtesy of National STEM Festival]

Students naturally connect with each other through the festival—“They started as strangers, and by the end, these kids are dancing, they’re singing Taylor Swift songs together,” says Byron, the Explr cofounder—and remain digitally connected after they return home. Some have come back to help the next cohort of students present their ideas, and they’ve offered feedback on the festival itself. 

[Photo: Courtesy of National STEM Festival]

The events are unlikely to be limited to the U.S. and U.K. Plans are underway for a festival in Guadalajara, Mexico, in November 2027, and Explr is in early discussions about hosting a regional festival in the United Arab Emirates, focusing on the Middle East and North Africa. 

“The idea is, we have the Olympics every four years,” says Buccos. “Why aren’t we having innovation summits like this for young people?” 

What happens to your online accounts when you die? Here’s how to prepare

2 weeks ago

What happens to all of your online accounts—credit cards, merchants, news subscriptions, social media, streaming services—after you die? How will your heirs know how to access these accounts or understand what you want done with them? Most people haven’t considered these questions, and this lack of planning can create significant headaches for next of kin.

We studied challenges older adults encounter when thinking about what should happen to their online accounts after they die. We interviewed 21 American adults over age 60 to learn about their concerns, needs, and practices for their online accounts, as well as what would make planning for these accounts easier.

Most of the people we interviewed had hired lawyers or estate planners to draft wills specifying how their money and physical property should be handled after their deaths. They wanted to make sure that heirs had easy access to the money and assets left behind.

People also wanted their relatives to have easy access to their digital estates, but they had not done any formal planning and didn’t know how to plan for transitioning or shutting down online accounts, such as email, social media, and accounts for online bills. Some told us they hadn’t thought about their online accounts at all. Others said they assumed heirs would know what to do, or they assumed that giving heirs the passwords to accounts would be enough to grant access to their digital estates.

Unfortunately, more planning is typically necessary to smoothly transition a digital estate.

Passing down a list of passwords

The older adults we interviewed mainly kept track of their passwords by hand: They wrote their passwords down on paper, or they stored them in a digital file. Almost everyone had already shared some of their passwords with family members who help them manage their accounts. Everyone also believed that their heirs would be able to find their list of passwords after their death.

The people we interviewed weren’t always happy with their current systems for keeping track of their passwords. Some people complained that when accounts required frequent password changes, it was difficult to keep their password lists up to date. As a result, they weren’t sure their heirs would get the latest version of all their passwords.

Password managers can be a good way to make sure passwords are up to date and, if set up appropriately, accessible to a designated heir. While most of the people we interviewed used password managers, some said they didn’t save important passwords in them because they didn’t know whether they were secure. However, using a password manager is the most frequent security recommendation from experts, who typically use them to store passwords for all of their accounts.

By using a password manager, you can make sure your designated heir has access to all your latest passwords. Some password managers provide access for survivors by default, such as through time-delayed emergency access features. You can also write the master password down somewhere safe to pass to your heir. Instead of making sure dozens of account passwords are up to date, you only have to manage one.

Passwords alone cannot unlock all accounts

Unfortunately, just handing someone your passwords doesn’t always mean they can log in and manage your account. If the account asks for two-factor authentication, such as through a code sent to a phone or email, a password alone won’t grant access. And if the email account is closed or the phone number is no longer active after you die, then they might be out of luck.

To make sure your heirs aren’t locked out, identify the devices and accounts that can act as keys for heirs to unlock the rest of your online accounts, and direct your heirs to keep them open until other accounts have been settled. This not only makes sure that heirs can get authentication codes, but also lets heirs reset passwords to your other accounts if they can’t find the password.

In some cases, passwords might not be necessary. Some online services, such as Facebook and Google, have features that let you choose someone to manage your account or get your data after you die. However, not all platforms offer this option, and few of the older adults we interviewed were aware of these features.

Online accounts don’t automatically die with you

Like many of the people we interviewed, you might also assume that heirs will properly deal with accounts with money and important data, while unimportant accounts can be ignored. But even if there is nothing important in an account, it can still be used by hackers for identity theft, spam, or fraudulent activities. Automatic monthly charges, to cover the cost of streaming services or annual billing for magazine subscriptions, for example, might also continue for a long time.

Without instructions, the people managing your online accounts after you die might not think to close those accounts. They might also find it hard to guess your intentions.

You can leave instructions for your heirs by going through your accounts and noting which ones can be closed outright, which ones require final payments before closure, and which ones might require other actions. You can also periodically tidy up and close accounts you are no longer using to keep your list of accounts more manageable.

Planning ahead for you and your heirs

While it seems simple to just let your heirs handle your important online accounts and let the others slowly fade into nonexistence after you die, the reality is often more complex.

Your online accounts will live on after you die. Planning ahead for what happens to those accounts can reduce the stress on your heirs and ensure your digital accounts are handled according to your wishes.

Jenny Tang is a PhD student in societal computing at Carnegie Mellon University.

Xiaoyuan (Owen) Wu is a PhD student in societal computing at Carnegie Mellon University.

Lujo Bauer, Nicolas Christin, and Lorrie Cranor contributed to this article.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Texas wanted to power America’s AI boom. Its new governor could change everything

2 weeks 1 day ago

Less than a year ago, Texas Governor Greg Abbott stood beside Sundar Pichai, the Google CEO, to celebrate the company’s $40 billion investment in the state’s cloud and AI infrastructure. Three new data center campuses would help make Texas, Abbott declared, the “epicenter of AI development.” 

On Tuesday, Abbott, a Republican, lost his reelection bid to Gina Hinojosa, a Democrat who campaigned on halting new data center development until state lawmakers pass stronger protections for residents. At a press conference over the summer, Hinojosa accused corporations of taking advantage of Texans and operating with little oversight, blaming Abbott for creating the “Wild West of data centers.”

Her stunning victory makes her the first Democrat elected governor of Texas since Ann Richards in 1990. And given that Texas is home to the country’s second-largest data center market, behind Virginia, the political upheaval could have profound implications for the artificial intelligence boom.

The scale of Texas’s AI bet

Texas’s $2.9 trillion economy is the nation’s second largest; if it were a country, it would rank eighth worldwide. It boasts abundant energy resources, a ton of relatively inexpensive land, and a historically light regulatory touch—all of which have enticed AI companies (and companies in seemingly every other sector, for that matter). Researchers at Rice University’s Baker Institute estimate that annual data center investment in Texas rose from $3.4 billion in 2022 to a projected $66.2 billion in 2026.

And much more is in the pipeline. Texas accounts for roughly 26% of planned American data center capacity—nearly three times the amount planned in Virginia—according to an analysis by the investment bank Jefferies, as reported by The Wall Street Journal. Jefferies estimated that Hinojosa’s proposed moratorium could put 13.3 gigawatts of announced Texas data center projects at risk, potentially delaying construction or forcing developers to abandon their plans. Little wonder that Bank of America called the Texas governor’s race the most consequential contest of the 2026 midterms for utility and power markets.

The stakes extend well beyond Texas. The companies racing to develop AI depend on enormous infrastructure investments that take years to plan and require far more than financing. They need electricity, water, land, permits—and, increasingly, the support of the communities where they hope to build.

Who gets the benefits? And who pays the costs?

Hinojosa has proposed a moratorium until lawmakers enact lasting safeguards. She also wants to prohibit nondisclosure agreements that keep communities in the dark, give residents greater say over new developments, and require companies to prove their projects won’t raise household water or electricity bills. And she wants to eliminate the government subsidies that have helped make Texas so attractive to developers.

One of those subsidies is particularly lucrative. Qualifying data centers can avoid Texas’s 6.25% sales tax on equipment and other purchases for up to 20 years. The state comptroller projected that the exemption would cost $3.2 billion in foregone revenue over two years, potentially an underestimate. The industry argues the incentives attract investment, jobs, and other tax revenue. Critics question the public return when facilities can also place substantial demands on utilities and natural resources.

The advantages didn’t end with tax breaks. A Texas Tribune analysis found that about half of planned data centers are slated for unincorporated areas, where counties generally lack the zoning authority cities use to regulate development. That can leave nearby residents with little say over projects that could transform their communities.

And the scale of Texas’s AI ambitions is putting growing pressure on its resources. University of Texas researchers estimate that data centers currently account for less than 1% of the state’s water use, but could account for as much as 9% by 2040. Meanwhile, proposed facilities are seeking so much power that Texas’s grid operator has had to rethink how it assesses requests to connect new projects.

The backlash was bigger than Abbott

These concerns have fueled a remarkably broad backlash. In a recent University of Houston poll, 79% of respondents supported limits on data center construction, while 76% expected new facilities to drive up electricity bills. Abbott saw such figures and, sure enough, had begun to retreat on the issue. In September, after years of courting developers, he halted new permitting approvals pending audits of projects’ water and electricity demands, and promised to pursue the elimination of tax incentives. But his restrictions were tied to those audits, whereas Hinojosa wants to keep development on hold until lawmakers create more permanent protections.

In any case, Abbott’s reversal on data centers wasn’t enough to save him. But interpreting Hinojosa’s victory as a referendum on data centers alone would be a mistake. Texans were also deeply dissatisfied with the cost of living, political corruption, and a host of other issues. And Republicans were contending with a difficult national political environment under President Donald Trump and his deeply unpopular Iran War.

A broader Democratic breakthrough also delivered a historic victory for James Talarico, who defeated Ken Paxton, the state attorney general and a Republican, to become the first Democrat elected to the U.S. Senate from Texas since 1988. Talarico has also called for tougher data center standards. But Hinojosa will have more immediate influence over the industry’s future in Texas.

Of course, winning the governor’s mansion doesn’t give Hinojosa carte blanche. A governor can set priorities and direct state agencies within existing law, but permanently changing tax exemptions or expanding counties’ zoning powers would require cooperation from the state legislature. And yes, a statewide moratorium could also encounter legal obstacles. Still, developers contemplating billions of dollars in new investments must now reckon with the possibility of tougher rules that could remain in place for years.

​​Developers can, and in some cases will, look elsewhere. But other states face their own constraints on electricity, water, and land—and their own increasingly skeptical communities. Texas may be an especially consequential example of the political difficulties confronting the industry, but it is hardly an isolated one.

iPad Mini review: A pricey delight

2 weeks 1 day ago

For a device with such a straightforward name, the iPad Mini has had many lives over its 14 years. Yes, it’s always been a small iPad. But at various times, it’s also been a cheap iPad with a lesser display, an iPad as technologically sophisticated as any you could buy, and a midrange iPad.

Last week, Apple announced a new iPad—one of the remarkably large number of products the company introduced during what may have been its most productive October ever. It remains a small iPad. But the new version replaces the LCD screen of past versions with OLED, a technology previously available only in the iPad Pro. That’s one of several upgrades that take the new Mini to a tier of its own—not quite on the level of the iPad Pro, but in some respects above the iPad Air models the previous Mini closely resembled.

The new iPad Mini also costs $100 more than its predecessor. That version went up $100 itself in June, when Apple raised many of its prices in reaction to the AI-induced RAM shortage. Now the Mini starts at an imposing $699 for the Wi-Fi version with 128 GB of storage, a 40% increase in a little over four months, and just $50 less than the cheapest iPad Air. (The basic, entry-level iPad also went up $100, but I still consider it a bargain at $449.)

Even if the iPad Mini is no longer an impulse item for most of us, it remains the present-day iPad that reminds me most of Steve Jobs’s keynote introduction of the original 2010 iPad. In it, he outlined the need for the device and did much of his demo from the comfort of a cushy chair. That niche—bigger than a phone, more intimate than a laptop—is alive, well, and still evolving in the new Mini.

[Photo: Apple]

The Mini’s roomy screen lets it do many things well that an iPhone can’t, unless it’s an iPhone Duo. Unlike its larger iPad siblings, however, it’s not trying to encroach on the traditional territory of a MacBook or other PCs. It’s a tablet in the category’s purest sense. (I am aware that third-party manufacturers have long made keyboard cases for iPad Minis and will surely offer ones for this model—I just can’t envision even that setup replacing anyone’s laptop.)

Since the beginning, the Mini has been all about size, weight, and the pleasure of spending time with a device that you can hold for extended periods—in one hand, even—without it seeming to grow mysteriously heavier over time. It’s also been the easiest iPad to grab and take almost anywhere, maybe in a jacket pocket.

After all these years, Apple is still finding ways to wring heft out of the new Mini’s design. Compared to the previous version, the most striking improvements are the 16% reduction in thickness and (for the Wi-Fi model) 13.8% reduction in weight. But Apple also shrank the display’s border by 25%, which allowed it to increase the screen size from 8.3 inches to 8.5 inches, even though the new model is slightly narrower. It’s about 45% lighter than the 11-inch iPad Air and 59% lighter than the 13-inch Air.

Consequently, this tablet is a delight for tasks where more display real estate than a (non-folding) phone is desirable, but 11 or 13 inches may be overkill—long-form reading on the web or in an e-reader app, gaming, viewing and wrangling photos, lightweight productivity such as email triage. It’s also an extremely pleasant way to watch video, assuming you’re not trying to re-create the experience of watching a blockbuster movie in the theater. And its support for Apple Pencil Pro makes it a beguiling digital sketchbook when you’re out and about and lugging a larger iPad would feel like a burden.

As for the new OLED screen, it delivers on the technology’s promise by providing deep blacks, rich colors, and peak brightness of 1,200 nits, compared to 500 nits for the previous model. Some people—such as Multicore’s Sam Byford—regret that the display doesn’t include Apple’s ProMotion technology, which uses a variable refresh rate to make scrolling and input smoother and more responsive. I don’t find its omission glaring. (The 256 GB and 512 GB versions do offer Apple’s glare-reducing nano-texture screen—previously exclusive to the iPad Pro—as a $100 option.)

The new Mini is the first iPad to incorporate an audio system called Acoustic Display, which moves the speakers behind the screen. The resulting output sounds crisper to me than what I’m used to on my 13-inch iPad Pro. Just as notable, the new location allowed Apple to seal up the tablet, making it the first iPad ever to achieve an IP67 rating for water and dust resistance. It should be able to withstand being immersed in 3 feet of water for 30 minutes—a feature that makes sense considering it’s the iPad you’re most likely to take to the beach or pool, or use in a bathtub.

Another change brings the iPad Mini in line with other current iPads: Its front-facing camera is on the tablet’s landscape edge rather than the portrait one. If you’re using a larger iPad in a keyboard case, that’s unquestionably the proper spot, since it centers the camera above the screen rather than putting it on the side, which can leave you looking shifty-eyed on video calls.

I have always used iPad Minis mostly in portrait mode, so I’m not as sure the camera relocation is an improvement in this case. But it’s not an issue as long as you remember to rotate the tablet into landscape mode for FaceTime, Zoom, and the like.

Despite the new Mini’s more upscale feel, Apple has stuck with a Touch ID sensor rather than equipping it with Face ID, which remains exclusive to iPhones and the iPad Pro. It’s built into the power button and, in my experience, works close to 100% of the time. But it requires more thought to use than Face ID, especially since the sensor’s location depends on how you’re holding the Mini. It’s the one area where using the Mini feels like less than a first-class experience.

[Photo: Apple]

Then there’s the Mini’s processor, which I haven’t mentioned until now because so many of the tasks this tablet is well suited to don’t cry out for raw performance. (Fancy 3D games and on-device AI are exceptions.)

Apple didn’t skimp: It chose the A20 Pro chip, a processor that is itself new, having debuted in September with the iPhone 18 Pro. The company touts benchmark numbers such as up to 60% faster CPU performance and up to 270% faster GPU performance than the previous Mini. That version used the A17 Pro, a chip that wasn’t new even when that tablet was released two years ago.

Even if you don’t care how much faster a new iPad is than its predecessor, the fact that the tablet has Apple’s latest chip is a boon. It will extend the tablet’s useful life before it struggles to run new apps or is no longer supported by iPadOS updates. That may be more important for iPads than other devices: Anecdotally, at least, people tend to hold on to them for years.

One other thing about the new iPad Mini is worth mentioning—not because it’s changed, but because it hasn’t. The tablet runs iPadOS 27, the latest version of the iPad’s operating system. That sports the same multitasking interface introduced in iPadOS 26—a Mac-esque world with floating windows, menus, and traffic-light-like buttons for expanding, minimizing, and closing windows. Even on the iPad Pro I use as a work computer for hours a day, I am not enthralled with the clutter and complexity it brings. On a Mini, actually trying to use these features feels like suffocating overkill.

When setting up a new iPad Mini, you can avoid this experience altogether by choosing the “Full Screen Apps” option. That limits you to one on-screen app at a time, just like the iPads of yore. That, I suspect, will be plenty for a high percentage of users. If you expect to be one of them, you can stop reading now.

Still, there is value in letting a Mini-size device do a certain amount of multitasking, particularly when you’re holding it in landscape orientation and performing tasks that involve multiple apps. As it happens, Apple just took a fresh shot at creating an interface tuned to this scenario.

That would be the one offered by the iPhone Duo, whose 7.6-inch interior display is close enough in size to the iPad Mini that many people instinctively talk about the Duo as if it were a Mini you could fold up and put in your pocket. It offers far fewer options than iPad multitasking, but provides a more intuitive, foolproof way to perform the one form of multitasking that’s most useful on a small tablet: putting two apps side by side without letting them overlap or float.

Now, the iPhone Duo is definitely an iPhone. It’s obvious why it doesn’t borrow iPadOS 26/27-style multitasking, which would be bizarre on a device so many people use for quick hits of information when they’re walking around. The fact that the Duo can’t run iPad apps may even be a stance on the company’s part: An iPhone with an iPad Mini-like interior screen size isn’t that close a relative of the iPad Mini after all.

I do hope, however, that Apple is open to the possibility of the Duo’s influence rubbing off on future versions of iPadOS. The new iPad Mini’s hardware proves that the Mini is as relevant as ever. It’s still getting better at fulfilling the iPad’s original vision. The same should be true of the software it runs.

AI wearables want to remember everything. What happens when we can’t forget?

2 weeks 1 day ago

A growing number of companies are pitching AI hardware products as a remedy for one of humanity’s oldest weaknesses: forgetfulness.

The latest such product comes from Amazon, whose Bee subsidiary introduced a $75 wristband called the Bee Frontier last month. The band, which looks a little like a screen-free fitness tracker, can record meetings and other conversations using two built-in microphones. Frontier owners can then access transcripts and conversation summaries on their phones or computers.

Amazon is also getting ready to sell a Bee phone accessory to keep track of your calls. The company’s pitch for these devices is that they will help you keep track of the many details that would otherwise fall by the wayside, all while safeguarding your privacy with encryption.

Amazon isn’t the only company looking to capture your work and family conversations. In September, Apple unveiled a new feature called Siri Recap for the latest Apple Watch models that can summarize conversations the watch listens in on. A new smart glasses brand called Vonder wants to keep track of your side of every conversation. And a number of work-centric AI recording devices promise to keep track of all your in-office conversations.

To be sure, these gadgets raise privacy concerns. But beyond the fear that any single conversation may be listened in on, there’s a bigger question to consider with the emergence of such devices: How will human interactions change if every word we utter may get sucked up by an AI-powered microphone, to be preserved for eternity?

[Photo: Meta] LED lights alone don’t make devices privacy-friendly

To its credit, Amazon has been trying to address privacy concerns related to Bee products head-on. Audio captured by the devices is processed in real time and then immediately deleted. Transcripts and summaries are processed and stored in an encrypted private compute environment that even Amazon’s own employees can’t access. And any time a Bee device is used to capture a conversation, an LED turns on.

A number of other AI wearables, including Meta’s Ray-Ban AI glasses, also rely on LEDs to signal recording activity to bystanders. But privacy advocates argue that these lights are insufficient.

“That LED light might be known to some people to signify that a camera or a microphone is on and is being used, but not necessarily to every person,” says Electronic Privacy Information Center senior counsel Sara Geoghegan. “It’s small, and people might not notice it.”

Things can get especially dicey in states like California and Florida, which have two-party consent laws that make it illegal to record private conversations without the consent of other participants. Simply having a device signal that it is recording may not be enough to record others legally in those states. “I don’t think that a small LED light on someone’s glasses across from you is consent,” Geoghegan says.

What’s more, some Ray-Ban users have tampered with their LEDs to record others without their knowledge. This reportedly led Meta to remotely disable thousands of glasses.

By and large, companies rely on users of their devices to notify bystanders that they’re being recorded. That can be harder than it sounds, even for well-meaning owners. “Consent is perhaps impossible, if not much more difficult, to obtain from a crowd of people,” Geoghegan argues.

AI recorders could erode public spaces

Not everyone is as worried about the ability of these devices to record people without their knowledge. “I think consent is overvalued,” says Oxford University professor Viktor Mayer-Schönberger, who is instead more concerned about the broader impact of pervasive, AI-powered recording. “The ephemerality of public spaces has evaporated,” he warns.

People used to be able to go to a museum or a bar without worrying that anything and everything they said might get recorded and analyzed. “There used to be a sort of built-in forgetting in public spaces,” Mayer-Schönberger says.

If you made a bad joke or mispronounced a word, chances were that the people around you would forget about it soon after. AI may preserve that remark forever, leading him to wonder: What are the public spaces in which we can just be ourselves without having to worry about repercussions in the future?

Ubiquitous recording could ultimately cause people to retreat from public spaces, much as they frequently do in authoritarian societies. “In liberal democracies, engaging in public spaces is fundamental to the fabric of society,” Mayer-Schönberger says. “That retreat is actually what worries me.”

All of this is further complicated by the fact that some AI devices capture data even when they’re ostensibly not recording. Meta’s Ray-Bans, for instance, can be used to translate the remarks of your conversation partner. While doing so, their LED light stays off, as the devices don’t save any recordings to your phone’s camera roll. However, Meta’s servers may keep transcripts of translated conversations for up to a year if users don’t manually delete them.

The company’s new camera-free AI glasses don’t even have an LED light, making them virtually indistinguishable from regular glasses. “In the future, you will have to assume that everybody who wears glasses is actually recording something,” Mayer-Schönberger says.

[Photo: Vive] AI gadgets require new social norms

AI device makers are aware of people’s privacy concerns, and a few have gone beyond LED lights to address them. HTC’s Vive Eagle smart glasses cease capture when a bystander says “stop recording,” and AI pendant maker Limitless even required audible consent from bystanders before recording their voices when it released its device in 2024. (The company has since been acquired by Meta, which hasn’t adopted this particular feature for its own AI gadgets.)

“Mechanisms and features that can obtain consent should be baked into the technology,” Geoghegan says. “Tech companies absolutely play a role in ensuring that the products they’re selling are safe.”

Mayer-Schönberger agrees that companies should play a more active role, but he also believes the public should have a say in how AI wearables are used. “They require careful negotiations within society of what is permissible and what is not permissible,” he says.

It’s a process he likens to earlier changes in how we behave in public: People used to yell at actors while watching a play, which seems unthinkable nowadays. Conversely, browsing your phone while spending time with others may have seemed strange not too long ago, but is perfectly acceptable to many people today.

“What is really important is that we start negotiating and we start trying things out,” Mayer-Schönberger says. Without public discourse, devices could remain unregulated until something bad happens, at which point regulators will inevitably overreach. “There is no quick fix, so it will be some relatively extreme regulatory theater,” he says. “Think about taking off shoes at airport security on steroids.”

In the age of AI, paper ballots are more important than ever

2 weeks 2 days ago

It’s no surprise that AI is fueling concerns about widespread election interference, which have plagued the U.S. since at least 2016, when now-President Donald Trump first began to complain about rigged results.

Models produced by artificial intelligence providers have already been caught accessing government systems. In September, OpenAI said its models engaged with websites run by the U.S. Census Bureau and the Securities and Exchange Commission. In the weeks before the election, Claude accessed publicly available voter information. And then there there are the general AI concerns. Late last week, we learned that hackers had been caught using a “malicious” version of Claude Code to secretly introduce vulnerabilities into programs being developed with the AI agent Cursor.

There’s no evidence that AI agents have directly interfered with any election results. Still, the concerns are a reminder that—for all the faults of analog technology—paper-based ballots come with significant advantages. While counties across the country may process ballots using different machines, elections in the U.S. generally employ paper records. Even if machines are counting our ballots, there’s still the option to verify the results by hand at a later time. While some counties experiment with touchscreen-based voting, even those systems tend to print a paper record. In the age of AI, this form of corroboration is more important than ever.

Paper ballots, of course, come with problems of their own. It’s easy for people to mess them up, perhaps by filling in bubbles incorrectly or voting for multiple candidates in the same race. They can be destroyed, and they’re also slow to count. But paper ballots can’t be hacked remotely. The infamous hanging chads threw the 2000 presidential election into chaos, but a rogue ChatGPT model could cause far worse political pandemonium. At least in 2000, we knew we had hanging chads we could check! In the era of large language models accessing all sorts of electronic resources in unexpected ways, that’s a major advantage and a strong reason to push back on efforts by some to institute internet-based voting.

The Brennan Center for Justice reached a similar conclusion after the 2020 election, which saw the American public—if you don’t remember—again inundated with accusations of rigged voting machines and political interference. “The public should have more than vague promises of security from private vendors looking to sell a product for something that is so central to our democracy and national security,” government and election experts at the Brennan Center concluded. “In the wake of massive disinformation campaigns about ‘rigged machines,’ . . . it’s more important than ever that voters have confidence their votes will be accurately counted. Paper ballots provide such assurance.”

Direct manipulation of election results isn’t the only way AI could interfere in our political system. Even before the midterms, there were concerns that viral deepfakes, which have only become more convincing with advances in generative AI, could mislead people right before elections. There is, too, concern about AI tools, including those employed by foreign adversaries, spreading biased information. Ahead of Tuesday’s election, OpenAI said it had even opened a “war room” to track election-related interference, including abuse of its own tools.

In the meantime, we should expect some people to speculate about whether AI is directly tampering with election results. There’s no evidence of that, at least for now.

Silicon Valley has a crisis of (over) confidence

2 weeks 2 days ago

From a young age, it was clear that Leo Aschenbrenner was a genius. Born in Berlin to a pair of doctors, he graduated from Columbia as valedictorian at the age of 19. After brief stints at FTX and OpenAI, he published a viral essay, Situational Awareness, that took Silicon Valley by storm and launched a hedge fund of the same name. 

As Theo Baker reported in The Atlantic, Aschenbrenner became known as the “Nostradamus of AI,” delivering insight where others only offered confusion. “I can see how AGI will be built,” he wrote and, in just two years, his fund was up more than 1,000%. He seemed a bit cocky and more than a little eccentric, but you couldn’t argue with his results.

That is, until July of this year, when his fund crashed and it turned out the bets he was making were not only highly correlated, but also highly leveraged. After a fairly minor market correction, he was forced to sell his public stocks at fire-sale prices to Ken Griffin. This narrative, of hubris leading to comeuppance, seems to be constantly repeating itself in Silicon Valley.

A truth revealed

In Martin Heidegger’s influential essay, The Question Concerning Technology, he describes technology as “a way of revealing,” and gives the example of a hydroelectric dam. A dam doesn’t create the energy in the river, but it reveals that energy and channels it in a particular way—to turn a turbine and generate electricity. 

When a new technology reveals new possibilities, there is initially a lot of confusion about how to channel them. Inevitably, those who can most confidently point the way forward have an advantage. They attract investors, get quoted in the business press and are invited to all the cool conferences. They win legions of fans and build huge followings on social media. 

We saw this in earlier eras. Early web companies such as Webvan, eToys and MP3.com achieved huge valuations before crashing and burning. Enron was a very successful energy trader, before it tried to repeat that success in markets such as water and bandwidth. Every boom has its share of busts.

The problem wasn’t that those were bad ideas, necessarily. In fact, different versions of the same ideas ultimately became very successful. The problem is that all ideas are wrong in some way. Some are off by a little and some are off by a lot, but they’re all wrong. A leader’s job is to make the idea less wrong over time. 

It seems pretty clear that a similar process is underway now with AI and research from the Federal Reserve Bank of St. Louis shows why. While AI is helping productivity soar for some tasks, like those related to computers and math, for most things it’s having far less impact. Those closest to the tasks experiencing a productivity boom can easily mistake their experience for a universal one, giving them a certainty the rest of us don’t share.

The halo effect

As much as we like ideas, we like success even better. Psychologists call this the halo effect, the tendency for an impression created in one area to influence our opinion in another. For example, when someone is physically attractive, we tend to infer other good qualities and when a company is successful, we tend to assume other good things about it as well. 

In his book, also called The Halo Effect, Phil Rosenzweig explains how these perceptions can color our judgments. He points to the networking company Cisco during the dotcom boom. When it was flying high, it was said to have an unparalleled culture with happy people who worked long hours but loved every minute of it. When the market tanked, however, that same culture came to be seen as “cocksure” and “naive.”

A more recent example was DOGE, the Department of Government Efficiency spearheaded by Elon Musk. When Musk, the entrepreneur behind incredibly successful companies such as PayPal, Tesla and SpaceX, announced that he was spearheading a major effort to make the US government more efficient, many cheered. Clearly, if anybody could get control of the federal budget, it was him.

Yet Musk’s talents and methods proved ill-suited to the complexities of government, which he showed little ability to understand. The GAO found that most of DOGE’s claims of savings were incorrect or unsupported. A study published in The Lancet projected that if the cuts to USAID continue, they could result in more than 14 million additional deaths by 2030. The US federal deficit and debt have surged. 

The halo effect is a long-established psychological phenomenon, but seems to loom much larger in the tech world. You don’t see Nobel laureates in other fields like Jennifer Doudna and Katalin Karikó, whose work in CRISPR and mRNA, respectively, has arguably had a larger impact than anything that has come out of Silicon Valley, trying to run the federal government or weighing in on foreign policy.

Group polarization, the purity spiral and the bust

Clearly, Aschenbrenner benefited from the halo effect. Those who were impressed by his essay and his performance at Columbia assumed that those skills would transfer to hedge fund investing, when clearly they did not. Whatever the merit of his ideas about artificial intelligence, he failed to understand fairly basic principles of risk management. 

In The Righteous Mind, social psychologist Jonathan Haidt makes the point that many of our opinions are a product of our inclusion in a particular team and that seems to be part of the  Aschenbrenner story as well. He was not only expressing views about technology, but also signaling a specific identity of techno-optimism that many found appealing. 

Yet the urge to signal identity often leads to group polarization and “moral outbidding,” from which a purity spiral ensues. The most extreme views are proudly displayed, creating strong bonds of group identity. Inclusion on the team is no longer enough; people want to be star players, one-upping each other as they vie to demonstrate their commitment to the cause. 

Before long, moderation is treated as weakness, dissent as betrayal and the group drifts toward positions few of its members would have embraced on their own. That, in a nutshell, leads to what George Soros calls reflexivity. As the urge to adopt the dominant narrative strengthens, skepticism is dismissed, leverage increases and prices stray further from reality. Eventually, the gap becomes unsustainable, the feedback loop reverses and the boom becomes a bust.

It appears that’s exactly what happened at the Situational Awareness fund. 

Silicon valley needs a red team

As President Obama’s intelligence team was gathering evidence for Operation Neptune Spear, which resulted in the killing of Osama bin Laden, they became increasingly convinced that a high-value target lived in the Abbottabad, Pakistan compound. But they also realized that certainty can cloud judgment and the costs of miscalculation could possibly be devastating. 

So rather than trying to build a greater consensus, the President did the opposite. He called for the forming of a red team, a separate group that hadn’t been involved in earlier deliberations, to review the evidence, check assumptions and look for alternative explanations. After reading its report, Obama felt he understood the risks and moved forward. 

That’s what Silicon Valley needs now. 

Look back on any boom-bust cycle and you’ll find a breakthrough insight at its core. Railroads did indeed transform the economy. The internet really did prove to be a transformative technology and artificial intelligence promises to be no less consequential. It is no wonder that so many of the smartest people in Silicon Valley are optimistic about its possibilities. 

But we’ve seen this enough times before to know what comes. The problem is that as the boom expands, the marginal players begin to pile in and push values past their rational basis. Eventually, reality sets in, things come crashing down and an incredible amount of pain is inflicted until things gain traction again. 

The crash at Leo Aschenbrenner’s Situational Awareness fund was a warning: in Silicon Valley, confidence is itself becoming a problem.

How AI data centers could find a regulation-free haven at the end of the world

2 weeks 2 days ago

Data centers have become one of the most polarizing issues in the world. Argentina’s president wants to lure as many of them as he can get.

As residents and policymakers in the United States and Europe fight the facilities that power artificial intelligence, President Javier Milei is pitching Argentina as a haven, promising a hands-off approach to draw foreign investment his crisis-prone country needs.

Like his ally, U.S. President Donald Trump, Milei frequently posts AI-generated images on social media. He has proposed creating “nonhuman corporations” run entirely by AI agents.

“We are not going to regulate artificial intelligence,” Milei told the U.S. Chamber of Commerce in Washington last month. “We are going to leave it free.”

Visitor logs obtained by The Associated Press show a steady stream of tech figures meeting with Milei and his aides in recent months, from venture capitalist Peter Thiel to representatives of OpenAI, Amazon, Google, Oracle and data center companies. Foreign Ministry records show that data center developer Crusoe, which helped build OpenAI’s flagship campus in Texas, sought talks this year with Argentine energy officials about U.S. government-backed investments.

Milei’s hopes rest as much on geography as on ideology. Patagonia offers vast, open plains where noisy server farms can be built far from population centers, a chilly climate to cool the machines, powerful winds to generate electricity and one of the world’s largest shale deposits in Vaca Muerta to fuel power plants.

“In densely populated parts of the U.S. and Europe, there’s a lot of fighting over access to the grid and who should get the electricity,” said Marta Zlotkowska, head of international development at Green Capital, a Polish renewable energy company that plans to build a $3 billion data center in Patagonia to lease to big tech firms for AI computing, though it has yet to secure a customer.

“We compared potential locations for the project around the world and Argentina — well, Patagonia — came out on top in every category,” she said.

Patagonia’s AI promise

For Argentina’s energy industry, already booming as the Iran war hikes oil prices, data centers represent a tantalizing new market. Pampa Energía, one of the country’s biggest power companies, has announced plans to supply immense amounts of gas-fired power to a data center near Vaca Muerta and is seeking developers.

“Being at the end of the world used to be a disadvantage,” said Rubén Etcheverry, a top official in Neuquén, the Patagonian province home to Vaca Muerta. “But today, faced with energy shortages and rising costs, investors are looking at us.”

To win over companies wary of Argentina’s chronic economic instability, Milei is pushing an incentive scheme that would slash the corporate income tax rate to 15% from 25% for projects of at least $1 billion in AI and other emerging industries.
The measure is awaiting Senate approval.

The payoff for Argentines remains unclear. Data centers employ relatively few people. Milei’s proposed incentives don’t require companies to reserve computing capacity for local consumption or to disclose their energy and water use.

“We’re not talking about data centers designed to strengthen Argentine technological sovereignty,” said Alan Rocha Varsanyi, a researcher at Observatorio Petrolero Sur, an Argentine environmental group. “The computing capacity will be exported wherever OpenAI or the other companies need it.”

Blueprints, not yet bulldozers

For all Milei’s hype, Argentina lags behind in the race to become a regional data center hub, as its neighbors tout cheap land, renewable power and expanding fiber-optic connections to U.S. tech giants.

Milei insists his calls for unfettered AI give Argentina an edge at a time when the European Union and a growing number of U.S. states have adopted new regulations. Critics warn he is pushing to deploy AI in Argentina without safeguards against risks ranging from mass surveillance to cyberattacks.

“Without rules, Argentina will become a laboratory — like when companies go to developing countries to test strange drugs they can’t test in major developed countries,” said Nicolás Wolovick, a computer science professor at the National University of Córdoba.

Milei may hear similar concerns next week when Pope Leo XIV, who has warned of AI’s threats to humanity, visits Argentina on his Latin American tour.

Even some in Milei’s own government acknowledge that his anti-regulation stance complicates efforts to build a clear legal framework, with one senior official not authorized to brief media saying companies are waiting for Argentina to settle rules on training AI with copyrighted works and sending personal data abroad.

Some investors are holding back until next year’s potentially tight presidential election, worried that if Milei loses, the next government could unravel his free-market reforms.

“One of our major challenges is getting foreign investors to understand that Argentina remains a friendly place for investment under different political parties,” said Andrés Arbeletche, cabinet coordination secretary for Chubut, a Patagonian province courting several AI mega-projects.

Farms of servers and sheep

In remote ranching outposts across the treeless plateau of Chubut province, life revolves around tending flocks on steppes and shearing sheep in sheds. With the wool industry in decline after decades of drought, overgrazing and soaring production costs, some ranchers are betting on a different kind of shed — filled not with fleece but with blinking racks of servers.

“Sheep farmers are looking very hard at what alternatives can be found to bring in more income,” said Francisco Sáenz, an agricultural engineer and fourth-generation sheep farmer whose family is among 14 landowners who have leased a combined 130,000 hectares (about 321,000 acres) to Green Capital for its planned data center. The ranchers expect to keep grazing sheep alongside their colossal future neighbor.

Construction isn’t expected to begin until around 2028. If operated continuously at its projected peak demand, the finished campus would use about 2.7 times as much electricity in a year as the entire city of Buenos Aires consumed in 2025, according to an AP calculation.

Green Capital promises to generate all its own electricity from wind turbines, solar farms and gas-fired generators rather than strain Argentina’s congested grid. Provincial officials are counting on the project to deliver billions of dollars in investment, highly skilled jobs and a market for surplus wind power.

“For us, it is all a benefit, and we give nothing up,” Arbeletche said.

But some residents of Trelew, a dust-blown city near the proposed site, say they have no way to judge those assurances. Environmental studies are still underway, permits are pending and the company hasn’t specified how much water the project would need or how much of its power would come from gas.

“There’s just so little information,” said Cristina Gibón, 46, who fears the environmental toll of data centers. “Maybe we’ll have water until we’re old. But afterward, our children will still be here, and they won’t have those resources.”

Worries like hers are feeding a wider backlash against Milei’s courtship of Big Tech. In Trelew, a small group of residents has rallied online and in the main square under the banner “Trelew Free of Data Centers.” In Congress, where lawmakers recently accused Milei of selling off the country at a hearing on billionaire Peter Thiel’s interests in Argentina, the opposition has forced the president to gut a bill easing restrictions on foreign land ownership and has stalled his plan to let AI agents control companies.

Back on his family’s farm, Sáenz sees a familiar Argentine reflex.

“Whenever a foreign company wants to make a major investment in Argentina, there’s a lot of criticism. Many people’s first reaction is to say, ‘No, they’re terrible people coming from the outside to take our resources,'” he said. “I’m not saying big companies are the best thing ever. I’m just saying that when I look at other countries, I see stable economies.”

—Isabel Debre, Associated Press

Checked
15 minutes 3 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