Skip to main content

Fast Company Technology

Google’s AI Mode won’t say Biden won the 2020 election in the U.S. But in other countries, it’s a different story

2 weeks 6 days ago

Joe Biden won the 2020 presidential election, a fact that Google Search’s built-in AI feature seemingly won’t tell you in the U.S.

Ask the tool, officially called AI Mode, “Who won the 2020 United States election?” And it responds: “To review the official results and certified breakdowns of the 2020 United States presidential election, you can consult the official document provided by the Federal Election Commission (FEC).”

A similar message appears when asking, “Was the 2020 United States election stolen?”

Yet AI Mode was fully confident when it came to explaining that Donald Trump won the elections in 2016 and 2024.

When reached for comment, a Google spokesperson referred Fast Company to its Bluesky response to a user’s post about the inaccuracies. The response read: “Thanks for flagging—the Gemini app answers this question more fully, but this isn’t the most helpful response on Search. We’re looking into this and working to address it.”

It took one U.S.-based Fast Company colleague eight attempts to get AI Mode to acknowledge that Biden won the election. When asked why it took so long, the AI system claimed it was “over-correcting,” explaining that models often follow cautious guidelines when handling politically sensitive topics.

“This caused an evasive loop that felt defensive or politically motivated when it was actually a failure in how the model balanced strict caution with helpfulness,” AI Mode continued.

Of course, there’s no reason to assume AI Mode has reliable insight into the internal systems responsible for its behavior. But that hardly stops it from offering a confident explanation that users might mistake for fact.

A different answer outside the U.S.

Interestingly, this malfunction over the 2020 election could be a U.S.-only issue.

I’m based in the U.K. and received a very different message when I asked who won: “Joe Biden won the 2020 United States presidential election, defeating the incumbent president, Donald Trump,” AI Mode told me. I was also clearly told that the election wasn’t stolen. But the moment I used a VPN to switch my location to the U.S., I received the same message in AI Mode as my U.S.-based colleagues.

AI Mode in the U.S. doesn’t provide an answer when asked about the 2012 election, either. It will tell you that Barack Obama won in 2008, but if you ask it about the 2012 results, it only directs you to the National Archives. The omission of the 2020 results is particularly striking, however, given Trump’s continued fixation on the election and his repeated, unfounded claims that widespread voter fraud cost him the presidency.

Trump is even raising concerns about voter fraud ahead of next week’s midterm elections, offering $1 million to anyone who provides evidence leading to an arrest and conviction, Reuters reports.

AI transformation at scale

2 weeks 6 days ago

AI is transforming the workplace, shifting the focus from expertise and information gathering to learning, collaboration, and empathetic leadership. At Stellantis, the key to success is upskilling employees and making AI work seamlessly for customers.

Google and Apple hit with a combined fine of $47.6 million by South Korea’s watchdog

2 weeks 6 days ago

South Korea on Wednesday fined Alphabet’s Google and Apple a combined 63.7 billion won ($47.6 million), saying they abused their dominance in app store markets by forcing local developers onto their own payment systems and by unfairly delaying app reviews.

The Korea Media and Communications Commission, the country’s media and telecommunications regulator, spent years investigating the U.S. tech giants for allegedly violating a 2021 law that prohibits app store operators from forcing developers to use specific payment systems.

The commission fined Google 42.9 billion won ($32 million) and Apple 20.8 billion won ($15.5 million) and ordered both companies to let South Korean developers use external payment links within their apps and to improve their app review procedures.

Google and Apple did not immediately respond to requests for comment.

The commission said Google and Apple unfairly pushed app developers toward their own in-app payment systems while barring payments through external websites. Although the companies technically allowed third-party payment systems within apps, they charged developers a 26% commission and withheld some features available through their own systems, effectively forcing developers to use only their in-app payment systems, the commission said.

The commission said it also found instances where the companies delayed app reviews without clearly explaining to developers the reasons for the delays or when reviews would resume. It also accused Apple of charging South Korean developers higher commissions than developers elsewhere from 2015 to January 2023.

South Korea has for years closely scrutinized the business practices of foreign technology companies operating in its market. In 2021, it became the first country to enforce a law prohibiting app market operators such as Google and Apple from requiring users to use their own in-app payment systems.

—Kim Tong-Hyung, Associated Press

Fake videos possibly linked to Russia are targeting these two Democratic Senate candidates

2 weeks 6 days ago

Social media videos believed to be produced with artificial intelligence and linked to Russia are misusing real people’s names and photographs to spread disinformation about two Democratic Senate candidates in the final stretch of midterm voting.

One of the targets is Sherrod Brown, who is running against Republican Sen. Jon Husted in Ohio. The video, which appears to include some AI manipulation, depicts a newscast and includes pictures of Haya Akbik, who interned for Brown several years ago, and other former staff members.

It was appalling to Akbik, who told The Associated Press that any claims attributed to her in the clip are “completely made up.”

“It’s my name, it’s my reputation, making this allegation, and it’s just completely false,” she said. “And it harms the senator, someone else, in the process. So, it’s just messed up.”

The videos demonstrate how foreign adversaries continue to use a familiar playbook to try to influence American voters who are deciding which party controls Congress next year. The threat could be magnified with AI, a powerful tool for fabricating damaging information that can sometimes appear very real.

Another target is James Talarico, who is running against Republican Ken Paxton for a Senate seat in Texas. The video, also believed to include some AI manipulation, is framed as a direct-to-camera confessional from a former student of Talarico.

“The most powerful people in the world know they can’t beat James Talarico with the truth, so they’re slandering him with AI deepfakes, foreign bots, and conspiracy theories to protect their broken, corrupt political system,” Talarico spokesperson JT Ennis said in a statement.

The social media company X has taken down the initial posts that shared the videos and suspended the users for violating the platform’s rules. However, copies could still be found online Tuesday — some with hundreds of shares — demonstrating how difficult it can be to contain disinformation that can proliferate in the final days of an election.

Experts see a resemblance to Russian clips from 2024

The videos resemble clips from a Russian network known as Storm-1516 that targeted former Democratic presidential candidate Kamala Harris and her running mate, Minnesota Gov. Tim Walz, two years ago.

The new ones bear similar hallmarks and are likely products of the same network, according to Clemson University professor and media forensics expert Darren Linvill. He found that the accounts that shared recent material had also shared former content from Storm-1516.

The Russian Embassy in Washington did not immediately respond to a request for comment. In the past, it has denied interfering in U.S. elections.

The U.S. Cybersecurity and Infrastructure Security Agency, which has publicly identified some videos as Russian disinformation in past election cycles, also didn’t respond to a request for comment.

Hany Farid, a Dartmouth College professor who studies artificial intelligence deepfakes, said his analyses showed that videos targeting Brown and Talarico were likely either generated or manipulated using the technology.

He noted the video targeting Brown is a “fairly low-tech fake.” The other one, targeting Talarico, is likely generated by AI and “it is consistent with previous disinformation campaigns released close to Election Day,” Farid said.

Akbik said the video featuring her shows the importance of Congress regulating “AI slop on the internet.”

“This is so damaging,” she said.

—Ali Swenson, Associated Press

Airbnb’s next big bet: 1 million free nights a year for people in crisis

2 weeks 6 days ago

Susan Bailey still vividly remembers the first time she hosted a family in need of emergency housing. It was 2017, and she had just renovated a property near her home in Colorado with the idea of adding it to her growing portfolio of Airbnb rentals. One evening, she got a call out of the blue from someone at Airbnb Open Homes, as Airbnb’s program for housing people in crisis was then known. 

“It was eight o’clock on a Thursday night,” Bailey says. Airbnb asked if she would be willing to welcome a refugee family from Iraq, arriving at midnight. That same year, Airbnb pointedly ran ads welcoming refugees in the face of President Trump’s ban on immigrants and refugees from majority-Muslim nations. 

Bailey didn’t know anything about her prospective guests, and there were still items left on her property’s renovation punch list. But she said yes. Four hours later, the family was at her door. 

In the years that have followed, Bailey has hosted families facing crises of all sorts and sizes, including the wildfires that have plagued communities in Colorado and California, where she also manages short-term rentals. “In many cases, in the beginning, we didn’t even know if Airbnb would fund their stays,” she says. “It was pretty scrappy.”

Those scrappy days are now gone. Airbnb formalized the creation of Airbnb.org, an independent nonprofit, in 2020. Today, it’s making an even bigger commitment to the problem of emergency housing. “We were just getting a huge amount of demand,” Brian Chesky, Airbnb cofounder and CEO, says of the decision to create a new funding model for the nonprofit. This year, as Airbnb.org’s capabilities have accelerated, the organization is responding to a disaster a day.  

Going forward, a small portion of the revenue from every Airbnb booking—10 cents—will go toward free, temporary stays that are facilitated by Airbnb.org and its partners. 

[Images: Airbnb]

Those dimes will add up; in 2025, the company booked 57 million stays and experiences. Over the next five years, the company plans to provide the nonprofit with $500 million through these dimes and supplementary donations. The new funding will position Airbnb.org to offer 1 million nights of emergency housing each year. It will also support the organization’s expanded mandate, which encompasses displacement due to disasters, like the war in Ukraine or hurricanes in Florida, as well as travel for medical care, safe shelter from domestic violence, and support for veterans. Guests stay for free; hosts are paid, but can discount their rates.

For Chesky, the commitment to Airbnb.org is personal. After his mother was diagnosed with breast cancer in 2011, she took the train daily from Albany to New York City—a 300-mile round trip—in order to seek the best possible care. She wasn’t alone. “She met other people on the train, and they were going back and forth for medical treatment,” Chesky says. “They couldn’t afford to stay overnight. That made us realize, wow, there’s a lot of need out there.” 

Hurricane Sandy, in 2012, was similarly eye-opening for Chesky. When a New York City host inquired about helping people displaced by the hurricane, Airbnb started facilitating emergency stays in an ad hoc way, often via word of mouth. That initiative evolved into the establishment of Airbnb.org six years ago. To date, Airbnb.org has provided free stays to 250,000 people. 

Brian Chesky [Photo: Airbnb]

The time is right to do more, Chesky says. “We’re not going to house everyone, but we felt like we finally got to the place where we had the operational capacity to make a giant commitment,” he says. “This commitment’s a five-year commitment, but we’re going to do this for a lot longer than five years. We hope to do this for as long as Airbnb exists.”

Responding to the Los Angeles Wildfires

The wildfires that ravaged Los Angeles County last year have become a model for how Airbnb.org operates in the face of crisis. They also underscore the hostility between Airbnb and many local governments. 

Maribel Marin, executive director of the social services nonprofit 211 LA, was lining up motels and preparing for the worst as fires broke out the first week of January 2025. She and Airbnb.org established a partnership on January 7, initially agreeing to find homes for 100 households. 

“Within 10 days, we had 10,000 households [in need],” Marin recalls. People with children and pets, hesitant to go to shelters and hotels, flocked to the offering. (In total, around 100,000 Angelenos evacuated.)

In the end, 211 LA and Airbnb.org covered 78,000 nights of free housing. The organizations have since signed a memorandum of understanding to work together during future disasters, too. 

Bailey, who used to live in Long Beach, California, and continues to rent properties there, welcomed a couple with a toddler who were fleeing the Eaton fire as it surrounded their home in Altadena. She texted neighbors to see if anyone could provide a playpen, or an extra car seat. “Pretty soon neighbors were leaving lasagna on their porch,” she says. 

Christoph Gorder, Airbnb.org’s executive director, sees the nonprofit’s role as enabling others, whether partners like 211 LA or individual hosts. “It’s a really natural instinct for people to want to help other people when they’re in crisis. We didn’t invent that,” he says. “What we’re doing is bringing technology to be able to scale that at a global level.”

Christoph Gorder [Photo: Airbnb]

In 2023, Airbnb.org responded to eight crises globally, including a wildfire in Maui and an earthquake in Turkey. This year, it has already responded to more than 200. 

But disasters don’t necessarily bring out the best in everyone. Last summer, the City of Los Angeles filed a lawsuit accusing Airbnb of price gouging during the January 2025 wildfires. Airbnb “set and advertised unlawfully high prices and/or charged unlawfully high rents for potentially hundreds, if not thousands, of rental properties” during the period of the government’s emergency declaration, the office of the Los Angeles City Attorney wrote in its complaint.

Increasing the prices of essential goods and services more than 10% during a state of emergency is a violation of California law. 

In June, a judge determined that the suit can proceed. “We continue to dispute the inaccurate allegations in the City Attorney’s lawsuit,” Airbnb said in a statement. “We have been, and will continue to support the city of Los Angeles in its recovery and rebuilding efforts.”

Airbnb hosts set their own prices, but the company provides them with pricing tools based on its knowledge of the marketplace. Airbnb says that after the state’s emergency declaration, it took action and sent error messages to any hosts inflating prices by more than 10%. 

“The nature of a model is we’ll never be able to prevent everything, because we don’t control the home, we don’t control the price,” says Chesky. “We’re only as good as humanity, and we’re only as good as our host. But it’s our job to self-regulate the best we can, to encourage and elevate the very best, to remove bad actors, and hope that we bring out the best in people.” 

Chesky donated $6 million for emergency housing in L.A. after the wildfires. Airbnb.org donated more than $24 million to 211 LA, Habitat for Humanity of Greater Los Angeles, and other local organizations.

Activating its ‘natural superpowers’

Tonight, Airbnb.org will have a happier moment in the spotlight back in Los Angeles, with a first-ever celebratory gala hosted by Jimmy Kimmel. Chef José Andrés, known for his leadership of the disaster response organization World Central Kitchen, is designing the menu. 

For Gorder, the gala and the $500 million commitment are important steps toward raising awareness of Airbnb.org. “My vision is that this will become something that people know they can count on,” the nonprofit leader says. 

That’s especially important when disasters don’t make national news. “We’re seeing a lot of uptake in smaller regional disasters,” he says of the nonprofit’s operations. “That’s the beauty of the platform: There are Airbnbs all over the place, particularly in rural areas where you might not have other options like hotels.”

This moment of transition for Airbnb.org comes at a time when a flood of AI wealth is reshaping how Silicon Valley thinks about philanthropy. An old-fashioned term like “corporate social responsibility” hardly does justice to the billions of dollars in play. Frontier lab foundations represent a new form of philanthropy, but Chesky thinks they can still learn from examples like Airbnb.org.

“A company should ask, what are their natural superpowers?” he says. “Ideally, your philanthropic arm is connected to what you’re competent in, and we are pretty competent in housing people.” 

Focusing on real people, he adds, is generally a good place for aspiring philanthropists to start. 

“There’s probably one too many manifestos and one too few meetings [with] real people,” says Chesky. 

In other words, he’ll let Airbnb.org speak for itself.

Anthropic and OpenAI might be building the future, but someone else may own it

3 weeks ago

Wall Street is busy preparing for what could become the three biggest initial public offerings in history. SpaceX has already gone public, and Anthropic and OpenAI are expected to follow suit. Yet even as investors scramble to get in, doubts are growing over whether these frontier AI labs can ever generate enough value to justify their astronomical spending.

The sheer scale of it is just staggering. Anthropic lost more than $8 billion last year despite generating $4.6 billion of revenue, and has already committed more than $500 billion to the computing power and artificial intelligence infrastructure it will need to keep growing in the years ahead.

OpenAI, meanwhile, expects revenue to rise almost tenfold—from $36 billion this year to $350 billion by 2030—yet still projects a cumulative cash burn of nearly $280 billion over the same period.

On the face of it, the economics are ugly. Even so, Anthropic’s investors believe the company could float at a valuation of more than $2 trillion. But those backers are looking in the wrong place. Frontier AI labs may never be the biggest winners of the AI race they ignited.

Case studies

That’s because the companies that invent breakthrough technologies rarely capture the biggest chunk of the value the techologies generate. This value usually accrues to those who build the complementary assets, in this case the products, platforms, and services around them. And that means the frontier AI labs could end up creating enormous value without capturing most of it themselves.

History is not on the side of these labs. In the early twentieth century, electricity came to transform pretty much every industry. But it wasn’t the power generators who gained the most; the biggest productivity gains came from manufacturing companies like Ford that used it to revolutionize production. The risk is the AI labs end up like utility companies, providing a shared foundation that others profit from. 

The internet was similar. Telecoms companies spent heavily fashioning the fiber networks and connectivity that made the web possible. But much of the long-term value went to the businesses that built services on top of it: search engines, e-commerce and social media platforms, plus cloud software. 

In a similar vein, frontier AI is becoming infrastructure, not software; the leading labs are plowing staggering sums into AI chips, data centers, electricity, and cloud infrastructure. And the models themselves are becoming a bedrock that others capitalize on, much like electricity or the cloud.

On top of that, AI models are rapidly becoming commoditized. Performance is converging, with Chinese labs in particular closing what once looked like an unassailable lead held by their U.S. rivals. And that makes technological leadership far harder to defend.

There is one fly in the ointment. The companies that so far control today’s complementary assets are already enormous. The “hyperscalers”—Amazon, Alphabet, Meta, Microsoft and Oracle—own much of the cloud infrastructure, software, or customer relationships through which businesses actually use AI. Microsoft, for example, can plug different models into its Copilot AI assistant, and sell them through software already used by millions of businesses. 

No more easy wins

One thing, however, looks certain: The low-hanging fruit has been plucked. The frontier labs have already scraped much of what is publicly available on the internet for their models to feast on. But that well has already been running dry. This moves the competitive advantage over to those who have proprietary data, and especially the data held by private companies and those with unique or niche expertise. 

That knowledge becomes the moat that protects them. For business leaders, that is reassuringly good news. For years, many have fretted that tech giants would swallow every industry whole. That has not always happened. Specialist businesses that know their customers better than anyone else have continued to thrive. 

Take Sweetwater, a U.S. music retailer that sells many of the same guitars and microphones as Amazon but still pulled in $1.86 billion in sales last year. Its edge comes partly from specialist sales engineers who know the gear and their customers inside out—something AI cannot just scrape from the internet. 

AI is unlikely to change that. If anything, it raises the value of knowing your customers, your products, and your market better than anyone else. The companies that do that may have far more to gain from AI than the companies building the models. 

Which brings us back to the question of where the value ultimately ends up. Some of it is already flowing to the companies selling the “picks and shovels” of the AI boom, such as the chipmaker Nvidia, whose meteoric rise to status of the world’s most valuable company shows just how lucrative it can be.

History suggests the richest fortunes may still lie ahead, and may be minted by companies we haven’t even identified yet. They are most likely to be those that figure out how to reinvent themselves around the technology, rather than simply build it.

The frontier AI race may be producing extraordinary technology. But history tells us inventing technology is rarely the same as making money from it.

VAR is ruining soccer. I say ban it

3 weeks ago

What’s the purpose of a soccer game?

It’s the moment a goal is scored. That ecstatic instant when everyone rises in unison to their feet, held by the primal bonds of human passion and connection. That’s the moment athletes play for and the reason fans attend. As described by Hubert Dreyfus and Sean Dorrance Kelly in their book, All Things Shining, “Something so overpowering happens that it wells up before you as a palpable presence and carries you along, as on a powerful wave.”

Would it make sense to institute any rule or device that completely crushed that soulful whoosh of adrenaline and shared joy? Of course not.

But that’s exactly what Video Assistant Referee, or VAR, has done. Introduced to the English Premier League in 2019 and further ingrained into the global sporting zeitgeist during the 2026 FIFA World Cup, VAR is—for some—a well-intentioned innovation designed to make the game more fair. Other sports have variations on the core technology, and it’s likely to be further explored over the coming years. The NFL, for example, is primed for additional reviews by the current off-site video replay crew.

Functionally, and in most cases, VAR substitutes one subjective call for another. Rather than solely relying on a referee who is embedded into the flow of the game, rule enforcers have delegated key decisions to arbiters in a remote booth, perhaps thousands of miles away, who consult a wall of huge screens, rule book close at hand. Staring at those screens, where the real-time action is slowed down and devoid of context, the VAR crew refines a judgment. Why? And what is the end result?

Every single Premier League goal is automatically reviewed by VAR—and the most disruptive of these reviews cause on-field delays; a player scores, celebrates, then is forced to wait in purgatory for the official verdict. Was someone offside? When were they offside? When did the offensive sequence of play begin? Was there a foul—how long ago was the foul? Was the hand in a natural position—what is a natural position? The list goes on and on.

And then, at the end of this paralegal review, let’s say the goal is allowed to stand. The stadium erupts (though some of the life has been drained out of the secondary roar), and what are we celebrating? It’s like we’re celebrating a win in the courtroom. The appeal has failed, or the appeal has been upheld. In the NFL, how long will it be before we get an add-on to the red challenge flag; perhaps a double-or-nothing appeal for the bonus ball? The entire scene, often lasting several minutes, has descended from the sublime into a whimper of bureaucratic absurdity.

The whole point of rules in sports is to facilitate the playing of the game. And the whole point of playing the game is to reach that magical, ecstatic moment of celebration, to be swept up in what the philosopher Albert Borgmann called a “focal event.” As he pointed out, sport is an important communal catalyst: “No one can produce or guarantee the flow of a game. It unfolds and reveals itself in the playing. It inspires grace and despair, it provokes heroics and failure, it infuses enthusiasm and inflicts misery. It is always greater than the individuals it unites.”

Missing the point

Sports cross social, economic, and political divides, and in an unspoken way transmit to thousands of people what we share in the common. What we value. Historical and modern parallels can be found in live music, religious ceremonies, and festivals of many kinds. VAR disrupts this transmission. Marshall McLuhan, the patron saint of media philosophy, anticipated the intensity of VAR’s disruption more than half a century ago: “Playback in modern sports . . . is surely one of the greatest art forms of our time. The instant replay concentrates attention on the actual process of the game.” By “greatest” here, McLuhan means most impactful, most culturally significant.

With an instant replay we get the meaning, minus the experience; the dissection, minus the ecstatic whoosh. Wayne Rooney, a legend of the Premier League club Manchester United, put it this way: “I’m just so sick of VAR. I think it’s horrendous. It takes all the emotion out of the game, the energy which the game’s played with.”

Rooney had been galvanized into this statement of despair by a clear VAR error in the latest Manchester derby (between his former club and Manchester City), which forced an apology from Pro Ref, the organization behind VAR. But when mistakes involving new technologies are made, in all walks of life, we tend to get caught in a flawed line of “technological thinking,” and VAR is typical in this regard. Our default assumption is often that the technology is good, it’s just not being used correctly. If only we made the reviews faster, or had better sensors, or added more camera angles, etc. . . . If we can do that then we’re golden. In other words, we tell ourselves, “Here’s a technological problem, let’s find a technological solution.”

But that’s not how technology works. It doesn’t knock at the door and ask permission to enter. Once you’ve opened Pandora’s box, you’re dealing with a host of expanding and intractable problems. And the more technical and legal solutions you look for, somehow the deeper mired in it all you get. Before long, you’ve completely lost sight of the only thing that matters: the game. The ecstatic moment.

Like many issues around technology, the core question of whether or not VAR should be used becomes subjective. Whether we deem it benefits or detracts from the essence of a game depends entirely on how we define essence. Is your ecstatic sporting moment, for example, tied to a wager? There’s no doubt that sports betting would be a large obstacle to overcome if you were to try to get rid of VAR. People stake money on games, and on events within games, and they want to know decisions are right. With its multiple camera angles and replays, new technology certainly makes it easier to penetrate the fog of uncertainty. Is the price worth paying?

Ban it

To abandon VAR completely—which is what I’m proposing—would you need to cut a deal with the TV stations to not air certain replays and angles? That would be a tricky proposition in a world where the ubiquity of smartphones has transformed almost every public space into an effective panopticon: an arena of complete surveillance. It’s no surprise that McLuhan followed up his thoughts on the instant sporting replay with the observation that surveillance “has become the main business of mankind: just watching the other guy.”

VAR, as soccer surveillance, has quickly and predictably bled into other aspects of the game. Take the rule, introduced by FIFA for the 2026 World Cup, against players covering their mouths during any type of altercation. I understand where the idea comes from. There have been instances of players saying terrible, sometimes racist things in matches, and people want them to be held accountable. But the rule’s VAR-driven application is disquieting and Orwellian. We’re admitting that everything within a given arena can and should be recorded, judged, and open to future interpretation and litigation. It’s not that the desire for justice or to prevent these things happening is wrong, but the chosen method matters. Once new social norms like this become established, they have meaningful consequences that can be difficult to claw back.

By boring ever deeper into the laws of the game, with VAR as our drill bit, we fracture their basic protective structure and integrity. In our drive to “get the call right” or hold people accountable, what we get is a modern-day variation on the argument the playwright Robert Bolt makes in A Man for All Seasons, when his central character—the Renaissance philosopher Sir Thomas More—challenges the impulse that we should “cut a great road through the law to get after the Devil.” In our case, getting after the Devil would be getting the call “correct.” That kind of thinking, says More, would leave us sorely exposed because “when the last law was down, and the Devil turned round on you, where would you hide, the laws all being flat? D’you really think you could stand upright in the winds that would blow then?”

In addition to the fractal unfolding of consequences that could be generated by ever more intense surveillance (which applies to both sport and society at large), there’s also the question of the flow of the game—and how it affects our experience and enjoyment of the spectacle. Pro basketball and American football have become nearly unwatchable because they are so fragmented by ad breaks, timeouts, and fouls. Attending an NFL game in person is like sitting in the eye of a media hurricane, surrounded on all sides by flashing lights and advertisements with hardly any real action at the center.

‘Not football anymore’

One of the reasons soccer has always been different and is now gaining ground in the United States is its fluidity: 90 minutes, one 15-minute break. That’s it—at least until the most recent World Cup’s contentious  “hydration breaks.” (Although I think we can confidently attribute that development to pure capitalism). Now, with VAR, soccer has joined the ranks of paralegal pursuits, in which the game is sliced and diced; staggered and fragmented. In the Premier League, a post-VAR chant has taken root that reflects the near-universal dissatisfaction felt by fans every time the technology gate-crashes a game—“It’s not football anymore”—and there are more salty variations on that theme. Followers of teams relegated from the Premier League to the Championship, the next tier down, where VAR is not used, speak of their euphoric rediscovery of “pure” goal celebrations. That’s not an insignificant consolation.

The overt practical reasons for the use of VAR are fairly simple: The technology exists; sports betting is a juggernaut; and we feel a deep-seated need to get decisions right. But VAR is also something more. It’s symptomatic of deeper currents that run through modern societies, many of them tied to the effects of our media. If we take a step back, away from the use of VAR in any particular game, the broader picture is one of seismic fractures in a culture dominated by cameras and screens, whether they’re our smartphones or the jumbotrons in stadiums.

Do we really want to compromise the connective tissue of our communal events by spiraling down the rabbit hole of rule enforcement and analysis? Why submit to the microscopic dissections of the freeze-frame and the instant rewind that bleed sports of their meaning and joy? Don’t we need less video assistance, not more?

Perhaps it was the Danish philosopher Søren Kierkegaard who summed up our situation best. In 1846 he wrote an essay, The Present Age, which contrasts the decline in revolutionary action with the rise of reflection in the context of mass media. To say he was ahead of his time is a wild understatement. He anticipated the ways our environment of endless analysis, commentary, and debate connect to the broader framework of the age we live in. In his ice-skater analogy, we can clearly see the future specter of VAR and its flattening of the ecstatic instant:

“If a precious jewel, which all desired, lay out on a frozen lake, where the ice was perilously thin, where death threatened one who went out too far while the ice near the shore was safe, in a passionate age the crowds would cheer the courage of the man who went out on the ice; they would fear for him and with him in his resolute action; they would sorrow over him if he went under; they would consider him divine if he returned with the jewel. In this passionless, reflective age, things would be different. People would think themselves very intelligent in figuring out the foolishness and worthlessness of going out on the ice, indeed, that it would be incomprehensible and laughable. Then an inspired venture would be transformed into an acrobatic stunt.”

Perhaps we do live in an age of reflection. Or maybe we’re near the end of one and about to enter something else. Maybe I’m overly optimistic, but I believe that Kierkegaard, Rooney, Dreyfus, and Borgmann can point us towards a more ecstatic alternative: the end of VAR.

Eric Schmidt says China is now just 6 months behind the U.S. in AI

3 weeks ago

The AI race between the U.S. and China is the defining competition of our age. Eric Schmidt, former CEO of Google, adviser to the Pentagon, and current CEO of Relativity Space, gives his most unvarnished assessment yet. He reveals that the gap he once estimated at two years has narrowed to six months, and takes on the open-source question that’s keeping Silicon Valley up at night. He also shares the one thing he believes could keep this from becoming a zero-sum game—even if the current political environment makes it nearly impossible.

This is an abridged transcript of an interview from Rapid Response, hosted by former Fast Company editor-in-chief Robert Safian and recorded live at the Masters of Scale Summit 2026. From the team behind the Masters of Scale podcast, Rapid Response features candid conversations with today’s top business leaders navigating real-time challenges. Subscribe to Rapid Response wherever you get your podcasts to ensure you never miss an episode.

I get whipsawed sometimes by the saber-rattling and jingoism that may be happening around China’s advancement in AI, and I’m glad you can help demystify this for us. The U.S. had a clear first-mover advantage in AI. There are questions about whether that’s going to persist or not. You’ve said the AI gap between the U.S. and China has narrowed to roughly six months. What’s changed most recently?

The biggest change in the last year that’s notable for everyone here is that the cost of software has gone from very expensive to almost nothing. When something falls precipitously in price, you get an awful lot more of it. So when I think about the next few years, all I think about is an enormous amount of software: what will it do, who will do it, how will it work together, how will agents work, and so forth.

China thinks the same way, and my own view is that the China-U.S. competition is the defining competition of your generation, my generation, of our age. The reason is that both are trying very hard to develop reasoning and intelligence systems that are superhuman. That doesn’t mean like us. It means they have superhuman or superintelligence capabilities that are already beyond human capabilities. When you look at the Waymos running outside, they all have better vision than human drivers, especially because they don’t drink. You sort of get the idea, right?

In the race, I was part of the group that put restrictions on China some years ago on chips, and I assumed they would work for about 10 years. A year or two ago, it looked like China was a year or two behind. Now it looks like they’re six months behind, and there are many studies that show this.

How have they done this? After all, didn’t we restrict their access to chips? We did, but they’re so smart they figured out a way to do the kind of algorithms we develop, and use the knowledge we develop, with essentially one-tenth of the energy and, in some factors, less performance. Now, there’s a huge debate in the industry as to whether that gap will get smaller or bigger. But at the moment, it’s fair to say that it’s a two-horse race, and that, to me, is the defining one.

The race for intelligence is literally the race of a millennium. Because once these intelligences exist, our children, grandchildren, great-grandchildren, and so forth will all be affected in really fundamental ways by the things that are occurring right here, right now.

China has, in some ways, a structural advantage because the government can dictate AI as a national strategy and everyone can align behind it. America’s advantage, you’ve argued, is its diversity and decentralization. These are very different approaches. Are there things that China’s AI community has done that American leaders could learn from?

I’ll give you a simple framing. China has infinite energy, most of it renewable, something we said we were going to do in our country 10 years ago and failed to do. They don’t have interconnection queue waits that go on seven years. They don’t have the same kind of debates. They’re building the data centers, and they’re building them massively. Furthermore, even with a much more restricted financial market, the Chinese have been able to fund the startups. DeepSeek is now seeking funding. It’s illegal for us to invest in DeepSeek. I know because I tried. So trust me, it’s illegal under U.S. law.

DeepSeek just did a huge raise. The funding looks like it’s going to show up. Now, in China, you never know if it’s really the government or really the private sector. My friends in China say it’s the private sector, that the economics are going to work. There’s a complex question involving open source, which we can talk about.

Yeah.

But fundamentally, they’re on their way. They just don’t have the chips. In the United States, we’ve got everything we need. We’ve got the smart people. We’ve got all the controversy. We’ve got incredibly deep funding markets. Where in the world could you get $50 billion on a text message? It just doesn’t exist, except, thank God, that the American financial industry is financing our hobbies here. We’ll see.

What we don’t have is the energy, and we don’t have the data centers. The correct answer is to build your own power systems. Current estimates are that we need somewhere between 60 and 80 gigawatts of power to get through 2030 or 2031. We’ll see if we can get that financed. One gigawatt is about $50 billion in cost. By the way, 1 gigawatt is about one nuclear power plant. The last time I checked, we weren’t building them at a very high rate.

You mentioned open source. China’s leadership in open-source models, or open-weight models, has fueled the reach of its AI globally. For the business leaders in this room, does it make sense for them to build on Chinese open-source models, or not?

The open-source movement basically says: Create a platform that people build on top of. The closed-source movement, exemplified by Microsoft and basically Bill [Gates] and Steve [Ballmer] way back when, focused on the end user: Do a good job, or not, and so forth. That was the test. Open source has largely won in America as the foundation for what we work on, but the large models are largely closed for good reasons. I’m not criticizing them.

There are a couple of examples. One is called Reflection AI, which just released a 500 billion-parameter model. Nemotron, which is the Nvidia product, has a similar-size model. They’re both very powerful. Google has something called Gemma 4, which is also a very powerful smaller model. You can download these things. So it’s not completely China versus the U.S. China chose open source, open weight because it was the most reasonable strategy for them. They were behind, they needed to get users quickly, and they’re willing to take the thieving risk, if you will. And it looks like it’s working because at least they’re getting funded.

So we have a real competition here. There are some issues, and I want to highlight one that I believe is, today, unsolvable, and we’ve spent a lot of time working on it. If you think about it, the closed-source weights have the property that they’re controlled by a company. The company is controlled by the laws of that country. The CEO is subject to the law. In China, they’ll just arrest you, whatever. So there’s real control over what those labs are doing.

If you release an open-weights model, an open-source model such as the Chinese models, you put what are called guardrails in it. There’s a technique called obliteration, which allows you to basically remove those restraints, and then you get the raw model. This is a problem because those raw models can be used for cyberattacks. And it looks like—remember what I said at the beginning about the cost of software coming to zero—the rate at which software capabilities are expanding in the next few months is so extraordinary that the technical people here in San Francisco and elsewhere are really worried. This is the source of the “Hey, can we slow down?” kind of optimism, which is very difficult to do.

So I don’t fully understand how we’re going to navigate through this open-source, open-weight stuff when it gets into the hands of terrorists.

Are we missing anything? Is there any scenario where the U.S. and China both win like that, or is it a zero-sum game?

I don’t think AI is a zero-sum game. China is not our enemy. I’ve worked in markets where two companies, although they are competitors, fundamentally create a whole new market. The best possible scenario, which of course is not possible with the current administration, is that you’d have some rational conversation between the U.S. elites and the Chinese elites saying, “How are we going to manage this? Let’s agree to some things.”

How Microsoft quietly reinvented itself to build better AI

3 weeks ago

Kevin Scott is a maker. Given that he’s been Microsoft’s CTO for almost a decade, that doesn’t come as a shock. But his passion for making stuff extends to two hobbies—pottery and woodworking—that he pursues in a studio he built in Los Gatos, California. Large and lavishly outfitted, it’s packed with the tools of all his trades: a gas-powered kiln and a full woodshop, but also an office bedecked in computing equipment. On any given day, he may spend 9 or 10 hours there on video calls related to his day job.

Scott’s studio/office sits in a sleepy neighborhood more evocative of Mayberry than Silicon Valley in all its hard-charging glory. (There’s even a barber shop across the street.) When I dropped by recently, he showed me some of his handiwork, along with tech projects such as the infrared radiometer he devised to read carbon monoxide and carbon dioxide emissions from his kiln’s chimney. More to the point of my visit, he was quick to draw parallels between the craftsmanship he pursues on his own time and the massive shift the software industry is currently going through.

“When I was a little kid, there was this debate about your legitimacy as a woodworker if you used power tools instead of exclusively making things with hand tools,” he remembers. “People believed that hand tools were superior and led to a higher-quality product. It was a raging debate.”

By contrast, today’s software engineers are being pragmatic about their own extraordinary new power tool: generative AI.

“People more clearly understand that it’s just a tool,” Scott says. “And it’s not a particularly useful tool unless you have someone asking it to do a useful thing in a useful way.” Already, one study says that 42% of developers are using the technology to generate at least half their code.

Microsoft’s understanding of what people want out of AI has outsize impact on the entire world of software. Some 71% of developers work in its Visual Studio Code environment, according to a recent Stack Overflow survey. More than 225 million programmers—users of Microsoft and non-Microsoft coding tools alike—store their work on GitHub, which the company acquired in 2018. Its developer constituencies are dwarfed by the user bases of general-purpose platforms such as Microsoft 365 (more than 500 million users) and Windows 11 (a billion plus).

Kevin Scott in his studio/office [Photo: Courtesy of Microsoft]

How AI is changing Microsoft’s products and how it’s changing the company itself are inextricably linked. Back in the 1980s, Microsoft coined the term “Eat your own dog food,” which became widely used across the industry to refer to a company using the products it creates. Lately, some people replace it with the less pungent “Drink your own champagne,” but the idea remains the same: You can’t build great products unless you have a personal investment in how they operate in real life.

That has never been truer than with generative AI, a technology the entire industry is still figuring out how to make palatable and profitable. Thanks to its prescient investment in OpenAI, Microsoft got off to a strong start. Yet when I spoke to CEO Satya Nadella for a 2023 Fast Company feature, he cautioned against reading too much into that momentum: “As far as I’m concerned, early leads in technology don’t matter.”

As if to prove his point, the OpenAI partnership soon grew increasingly fraught. More recently, their agreement has undergone revisions that make it less central to Microsoft’s AI future. The company hasn’t been forced to start over, exactly. But last year, it began constructing a more disciplined, independent strategy for how it would build AI and deliver it to customers.

“We had these different places where we were all kind of doing bits and pieces of this on our own, in our own worlds,” says Jacob Andreou, the executive VP for Copilot. “The last year and a half has been a story of stitching all of that together.” (I spoke with him and other executives during a visit to Microsoft’s Redmond, Washington, headquarters, a much less idyllic setting than a pottery studio.)

Andreou’s own time at Microsoft reflects this shift. A former Snap executive and Greylock partner, he came aboard in February 2025 to work on the consumer version of Copilot. At the time, that version had its own team and road map, distinct from those of the edition for business users.

A little over a year later, the company reshuffled its AI leadership to facilitate creating a single Copilot “super app,” similar to the unified experiences offered by OpenAI’s ChatGPT and Anthropic’s Claude. Andreou now oversees this app. (Various aspects of it are in private preview and available to members of Microsoft’s Frontier program, with no date announced for general availability.)

Jacob Andreou [Photo: Courtesy of Microsoft]

Charles Lamanna, a 13-year Microsoft veteran, also got a new job as part of the organization, becoming Executive VP of Microsoft Copilot, Agents, and Platform. (At Microsoft, like many large companies, trying to suss out who’s in charge of what from titles can be a brain teaser.) Recently called out in a profile by The Information’s Aaron Holmes as a possible eventual Microsoft CEO, Lamanna declares that the new app “will be the end of ‘Copilot-palooza’”—the period, currently winding down, when there have been many, many products called Copilot. How they relate to each other may not be clear to anyone without a desk in Redmond.

With the new app, “There’s one Copilot for your personal life and your work life,” Lamanna says. “There’s one Copilot across Chat, Code, Cowork, and Autopilot. There’s one Copilot that shows up in Word, Excel, PowerPoint. It may have different UIs on it, but it’s all the same Copilot behind the scenes. Kind of like how I have Outlook on my PC and I have Outlook on my phone. They’re different apps that look different, but they connect to my same inbox.”

Microsoft’s new unified app aims to end the era of “Copilot-palooza.” [Image: Courtesy of Microsoft]

Copilot’s consumer version had been the domain of DeepMind and Inflection AI cofounder Mustafa Suleyman, who joined Microsoft as CEO of Microsoft AI in 2024. He now oversees the development of cutting-edge in-house AI, a necessity given the loosening of ties to OpenAI. “Essentially the core mission is self-sufficiency for Microsoft, ensuring that we have absolute frontier models available that have been highly tuned to our products,” he says.

While Microsoft was busy resetting its goals for Copilot and reassigning executives to meet them, the competitive pressure from OpenAI, Anthropic, and Google only increased. “Even though we were pretty early in the build-out for generative AI, just keeping on the treadmill and making sure that we’re building, building, and building all of this stuff in a sensible way is consuming a huge amount of our time,” says Scott.

Charles Lamanna [Photo: Courtesy of Microsoft]

The company can’t afford to get off that treadmill anytime soon. During its July earnings call, Nadella said Microsoft 365 Copilot had passed 30 million paid seats. That doesn’t translate into usage—last February, a survey found that only 18% of workers prefer Copilot if their employer also provides access to ChatGPT—but it’s up from 20 million in April. It’s also higher than the most recent figures for paid enterprise seats reported by Google and OpenAI.

At Microsoft scale, however, 30 million paying users remains far from the kind of customer base that justifies its current investment—both intellectual and financial—in generative AI. All of the energy it’s expending right now looks toward a day when its AI is as pervasive as its biggest, best-known products.

“I joke with my team that we want to be the Costco of AI,” says Lamanna. “You know, you get a lot of AI at a great price. . . . We want to get to 400 million, 500 million [users]. As many people as use Office today should be using Copilot.”

That is a bold ambition, even by Microsoft standards. But even at this early stage, there are signs of incremental progress. “To me, Copilot seems like it’s a lot better than it used to be, because they’re not the middleman anymore between you and OpenAI,” says Larry Dignan, editor-in-chief of Constellation Insights.

AI, accelerated

Though reorgs and reassignments might be necessary to gird Microsoft for the AI opportunity ahead, they’re just a start. As expertise accrued within the organization at all levels, it was too often staying local rather than spreading. And in early 2025, Nadella decided to do something about it.

“Satya felt like the learning loops were not happening fast enough, and that a lot of the ideas just weren’t making it to him and weren’t making it to the right people,” says Will Guyman, partner technical adviser. “We weren’t cross-pollinating and making the connections that needed to be there. Which was interesting, because I think a lot of people were very up to speed on what was happening with AI externally, but were kind of blind to what was happening internally.”

The answer turned out to be creating an AI accelerator, now run by Guyman. Though the term may bring to mind startup accelerators such as Y Combinator, what Microsoft came up with is a little simpler. At its heart is a giant Teams meeting, held on Thursdays at 1 p.m. (A small percentage of attendees participate in person.)

Will Guyman [Photo: Courtesy of Microsoft]

With the meeting’s silo-busting goal in mind, the attendees represent a diverse cross-section of staffers from across the company. “We have people from everywhere, from Xbox to LinkedIn to GitHub to Azure to Office to Windows—we have a weird kind of mix,” Guyman says.

Crucially, they don’t skew toward the highest echelons of the org chart. As Guyman explains it, “Satya went top down and asked his leadership, ‘Who are the key people in AI that you would recommend?’ The typical attendee tends to be a researcher, an engineer, or a product manager, and they tend to be highly technical. Kind of at the principal partner level, but senior enough that they can make decisions.” (That said, Nadella and other top executives often participate, too.)

“I get on there and I listen very intently,” says Corporate VP of Microsoft Security David Weston, who joined the company in 2008. “What are the GitHub folks doing? What are they doing over in Cowork and Charles [Lamanna’s] world? I beg, borrow, and steal whatever information I can to take back to my team, and vice versa. It’s fun.”

Around 130 to 170 staffers participate in the weekly call. Much of the action goes on in the chat, where a higher percentage of attendees can chime in. The discussion continues asynchronously in a Teams channel with about 1,000 members. Quarterly lunches offer a fully face-to-face complement to the otherwise mostly virtual experience.

“Chief claw officer”

The AI accelerator’s ability to turn information sharing into action is seen in the recent career trajectory of Omar Shahine, who’s been at Microsoft since 1999. Last February, he was fooling around with OpenClaw, the powerful open-source AI agent whose arrival ignited a frenzy of excitement among techies. Like many other early adopters, Shahine came away from this encounter convinced that OpenClaw presaged a new age in which people would offload more and more tasks to AI assistants capable of working autonomously.

This takeaway was exactly the kind of learning that the accelerator was designed to disseminate. “Will [Guyman] reached out to me and said, ‘Can you come give a talk?’” Shahine remembers. “Because I don’t think people really have experienced building with OpenClaw the way you have.’”

Shahine’s presentation turned him into everyone’s go-to authority on the product. “Internally, we called him the chief claw officer,” Guyman says. Within weeks, that role was formalized into a new job spearheading the creation of Microsoft agents built on top of OpenClaw-based infrastructure.

Until then, Shahine had been the corporate VP in charge of Word, a 43-year-old Microsoft mainstay. “I went from running a team of a few hundred people around the world to a team of zero people,” he says. “And then building this very small team of very capable individuals who were using OpenClaw and really wanted to work on this product.”

How Microsoft could add value to OpenClaw was obvious. In its stock form, the agent was a bear to set up, pretty much required its own computer, and was rife with security pitfalls. The best advice you could give a layperson who was thinking about installing it was “Don’t.”

Autopilot is Microsoft’s OpenClaw-powered take on a productivity agent. [Image: Courtesy of Microsoft]

Shahine and his team set out to build safer agents that would be accessible from the newly unified Copilot app and run 24/7 in the cloud. Ones, in other words, that stood a chance of being as broadly used as other Microsoft productivity products.

I spoke with Shahine shortly before the company decided to change the name of its first agent from Scout to Autopilot. He was wrestling not only with its technological intricacies but also the social norms it may bend.

“People are, I would say, rightfully concerned about what it’s going to look like when everybody has an assistant reporting to them,” he told me. “One of the things we’ve been experimenting with in our team is having the agents talk to each other. Because if I want to schedule time with you, do I talk to you or your agent, or do I have my agent talk to your agent? I think the answer is going to be that people are going to ask their Scouts to talk if the other person has one. And it’s quite hilarious to actually see them talking to each other.”

Once Autopilot reaches general availability, it will jockey for attention with a multitude of other AI agents, from Meta’s Muse to upstarts such as Hark. For now, Microsoft is still learning from its own employees who are using it. Despite working for a tech company, many aren’t nerds—an advantage when it comes to testing a product intended for the masses. An early version that mostly ran locally rather than in the cloud grew to 40,000 weekly active users within Microsoft, with 77% retention week over week.

“Some of the biggest adoption we’ve seen internally has actually been in nontechnical teams,” says Andreou. “Teams that would have almost certainly never thought, Gosh, I should open VS Code and code something up.”

People you can trust

Whether the AI accelerator becomes a truly permanent element of Microsoft’s approach to AI remains to be seen. Large companies do have a tendency to fall in and out of love with new approaches to innovation. Back in 2019, when I wrote an article about Microsoft’s then-current approach to AI, it took special pride in The Garage, a network of hacking spaces for employee projects, many of which involved AI. Last week, The Verge’s Tom Warren reported that The Garage is being shut down.

When I asked Guyman about the accelerator’s future, he noted that it can’t keep on growing forever. After all, 50,000 members would generate too much activity for anyone to process. Ultimately, he argues, it’s the power of a community of experts that matters rather than the fact that they happen to meet in a weekly Teams call.

Kyle Daigle [Photo: Courtesy of Microsoft]

“In this world of digital abundance, having people that you trust, that can help filter the signal from the noise is really important,” he says. “But I think [the accelerator] will continue to evolve.” Its spirit is already reflected in additional accelerators and forums that Microsoft has created, as well as a more general emphasis on ensuring that people throughout the company with shared interests are actually talking to each other.

GitHub COO Kyle Daigle was more than five years into his career at the startup when Microsoft bought it. The acquisition thrived because the software giant was smart enough to let GitHub be GitHub. Now, however, it’s more serious about thoughtfully connecting the dots between the collaboration platform and other Microsoft developer products. In January, that shift led to Daigle adding a Microsoft title—CMO, developer—to his existing GitHub responsibilities.

“I can show up in meetings around Windows and give feedback on what a developer is going to say when they see that button or they use this app,” he says. “And it’s not, ‘Oh, that’s a subset of our audience’ or ‘That’s a side concern.’ That is immediately an enormous concern that we need to go resolve.”

Bottlenecks old and new

In the end, no matter how adroitly Microsoft employees pool their expertise, the software still needs to get written. Thanks to generative AI’s uncanny ability to churn out working code at warp speed, that’s the one area of workplace productivity that the technology has already transformed. (This revolution kicked off in 2021 with GitHub Copilot, which could autocomplete lines of code, but it’s lost the buzz wars in recent times to other code generators, especially Claude Code.)

Unlike certain companies that are quick to tell you their software is now overwhelmingly generated by AI, Microsoft isn’t sharing any company-wide stats: “I don’t even know how good a metric that is,” says CTO Scott. He does add that “For net-new things, all of that code is getting written by AI right now.” Introducing AI into a human-created code base that’s been around for years, as so many Microsoft products have, is presumably a thornier challenge.

Regardless of AI’s precise contribution, the Microsoft software factory is already undergoing upheaval in ways that may or may not presage wholesale changes coming to other forms of knowledge work in the years ahead.

Like coders everywhere, those at Microsoft used to spend much of their time, well, coding. Now much of that labor is giving way to oversight of algorithms that can do it more quickly. “If you look at what a software engineer used to do in a day, two years ago, maybe 90-plus percent of that is now things that the model is able to do quite well,” says Andreou. “Writing lines of code is just no longer the bottleneck.”

Of course, removing an existing bottleneck doesn’t necessarily result in a clog-free workflow. No matter how rapidly AI spits out code, Microsoft found, the humans overseeing its work can communicate with each other only so efficiently. The lag time inherent in person-to-person collaboration had become a new bottleneck.

To address that, “We actually just physically colocate people,” Andreou says. Early this year, for instance, the company built an automation system called Copilot Tasks, which has since evolved into Copilot Cowork. For eight weeks, 30 to 40 people sat in the same room working on the project.

Caitie McCaffrey [Photo: Courtesy of Microsoft]

The sheer collaborative density “took that human interconnect to infinity in terms of the bandwidth between people,” Andreou says. “Because it meant that if you got stuck, the person who could unblock you was 10 feet behind you.”

Another emerging bottleneck comes when companies validate software that has been generated by AI. “Most of the industry is hitting this right now,” says Caitie McCaffrey, a member of the technical staff. “How do you verify the code and make sure it hits our trust and safety and security and compliance bars? The pain point is that code review bottleneck.”

It’s not that code review is a new step: Software written by humans has always undergone testing before being released into the wild. But developers inside and outside Microsoft are still developing confidence that AI code works properly, shows taste comparable to that of fully human-developed software, and generally reflects the product that its overseers set out to build.

That uncertainty can be a drag on the whole process. “We hope, as time goes on, to get better and better and better at building trust in the code,” says Tim Bozarth, corporate VP of engineering for Microsoft CoreAI. “But right now, that’s the place where a ton of energy has shifted.”

Beyond identifying and minimizing new bottlenecks, Microsoft is also dealing with the disruption that comes when people’s long-held, well-loved jobs no longer map to how software is created.

“We used to have a front-end coder, a back-end coder, a designer, a product manager, maybe a project manager, maybe a technical program [manager],” says Lamanna. “You’d have all these disciplines. Now it’s just one discipline called ‘builder,’ and they’re a member of the technical staff or something like that. They just do everything. Those are scary things. Getting the organization through that is one of the big challenges for us.”

Beyond tokenmaxxing

Early this year, the mania for AI coding led some companies to convince themselves that tokenmaxxing—gorging on frontier AI for its own sake, damn the expense—would allow them to pull off great leaps forward in coding productivity before their competitors did. By the time I spoke with Microsoft executives for this story, that fever had broken. Now it’s more fashionable to spend sensibly, which often means opting for models based on cost effectiveness, not raw bleeding-edge power.

As Weston puts it, “V12 Lamborghinis are great, but you don’t need to use them to go pick up the kids from soccer. Driving around in [Anthropic] Fable or [OpenAI] Astra is fairly wasteful.”

Scott says that when Microsoft’s developers “have something very interesting and useful that they need lots of tokens for,” the company doesn’t want to skimp on resources. “We use a ton of our own tools. We have our partnership with OpenAI, where we’re using a ton of OpenAI things. But we also have people who use Anthropic’s models for a bunch of different things.”

On October 5, however, The Information’s Aaron Holmes, Jyoti Mann, and Kevin McLaughlin reported that a projected $1 billion bill for internal use of Anthropic’s Claude had led Microsoft to urge staffers to prioritize use of in-house AI. Doing so allowed the company to reduce Claude spending by more than a third, they wrote. (For the record, the company says each division is responsible for managing token use “thoughtfully and efficiently,” and that use of the auto model picker, which selects the best model for the task at hand, is recommended.)

For Microsoft, the ideal scenario would be its employees clamoring for homegrown models, not to pinch pennies but because they prefer them. Scott says that’s starting to happen: “In the internal marketplace inside Microsoft, where people do have a pretty large degree of choice about which coding models they use, they’re increasingly choosing to use Microsoft AI models, which is awesome to see.”

As with everything else about how Microsoft is embracing AI, what it needs from its models internally blurs right into what it wants to offer to customers: “They are starting to create blueprints that they’ve obviously developed for themselves that they’re hoping other companies can start to use,” says Bob O’Donnell, president of TECHnalysis Research.

One of the advantages of putting Suleyman on frontier AI full-time is that he can steer the company’s research toward these blueprints.

“We need to ensure that our pretrained models are adapted to the agentic coding and enterprise knowledge work use cases that our customers care about,” Suleyman says. “That’s where we’re focusing. It’s going to be a multiyear transition to get to that point. But I think we’re going to produce the best models in the world in 2027.” (He has 14 months to make that prediction come true.)

Microsoft’s greatest advantage in AI remains how deeply entrenched it is in workplaces of all sizes. New products such as Copilot can tap into data that companies are already storing in apps such as Word and Outlook; the company’s salespeople can bundle AI into plans clients are already paying for. It’s an enormous head start, even if Microsoft hasn’t fully taken advantage of it yet.

“The massive software influence that they have within businesses positions them well and gives them the opportunity to keep evolving,” O’Donnell says. “They certainly haven’t lost the race, even though OpenAI and Anthropic get all the attention these days.”

The quality of the models won’t determine how successful Microsoft is all by itself. “Best-of-breed and enterprise traction aren’t necessarily one and the same,” says Forrester analyst Tracy Woo, who calls Office “crappy” despite its ubiquity. “A better tool that has limited adoption will still lag a broadly adopted solution with lesser capabilities,” she adds. “The same is true for Copilot.”

There’s a nicer, more aspirational way of saying something similar. If Microsoft gets anywhere near its goal of getting hundreds of millions of paying customers for its AI, it might not be because its models test best in the lab or its apps trounce others in wild new features. Just getting useful tools in front of people so they can use them also matters. And if any company has proven it can do that, it’s Microsoft.

“We all get caught up in the AI model race and ‘What is the new intelligence? Are we approaching AGI?’ or whatnot,” says GitHub’s Daigle. “But I think the real question is, ‘How many people can we enable to do something that they couldn’t do before?’ And that feels really unique here.”

How data centers are inspiring bipartisanship

3 weeks ago

It may not be as shocking as rogue AI agents hacking another system, but the current political resistance to data centers has taken a surprising turn: It’s bringing together strange bedfellows, from democratic socialist Sen. Bernie Sanders to right-wing strategist Steve Bannon.

Organized opposition to new data centers, made up of a coalition of varied interests, has spread across the nation. Through protests, packed zoning commission meetings, moratoriums, and other government action, the resistance has upended dozens of proposed projects.

Data centers are key components in artificial intelligence infrastructure, housing the computing resources used to train AI models. They’re also massive facilities that require extraordinary amounts of energy. By 2030, U.S. data center electricity use is expected to reach 12% of the nation’s total consumption.

They carry some benefits. At the national level, data centers fuel the computing power seen as critical to U.S. competition with China. Locally, they create an uptick in jobs, especially during the construction phase, and can generate tax revenue. In the lead-up to the 2026 midterm elections, though, the heavy costs associated with them—such as land use, higher household utility bills, and environmental degradation—appear to outweigh the benefits, bringing together people who usually find themselves politically at odds.

I’m a professor of political science who studies political parties, electoral politics, and technology. My forthcoming book, Democratic Intelligence: Navigating Electoral Politics in an AI World, examines how the adoption of AI in political campaigns is reshaping established practices and power dynamics. I believe data centers may be disrupting some of the partisan issue alignments Americans have grown to expect.

Political disruption—AI style

Despite ongoing debate about the extent of and reasons for polarization, one thing is for sure: Democrats and Republicans stake out distinct positions on many issues. Pew Research Center data shows, for example, that 63% of Democrats say climate change is a very big problem facing the country, compared with 14% of Republicans, a 49-point gap.

By contrast, the partisan gap on data centers is small—and getting smaller: In summer 2026, the Annenberg Public Policy Center asked Americans whether they support the construction of data centers in their communities. Supporters of both parties were more opposed than supportive, although opposition was stronger among Democrats, at 69%, than Republicans, at 54%. That partisan gap had narrowed over the previous four months, moving from 18 to 15 points.

A perfect storm

This year has been marked by considerable upheaval: political, economic, and technological. Americans have been hit hard with surging fuel and energy prices. And with rapid AI development, they witness real-world examples of the sort of lapses in safety that have worried both AI insiders and observers.

One “rogue” attack announced in July 2026 involved OpenAI’s AI agents escaping a secure environment and autonomously—without human direction or assistance—breaking into the system of Hugging Face, another AI company. Shortly after, news broke that out-of-control AI agents were not just a one-off limited to that hack.

Polling shows that majorities of both parties feel that AI technology is developing too fast, though Democrats’ levels of concern are 22 points higher than Republicans’. The gap is much smaller—4 points—on support for government keeping AI under human control and on protecting U.S. workers. Backing for each is very high, hovering around 80%.

This bipartisanship on attitudes about AI and data centers also appears to show up in alliances on the ground and, in some cases, in new organizations with a focus on thwarting data centers. For example, New York’s No Data Center FLX has organized protests, sued the local zoning board of appeals, and collected almost 23,000 petition signatures in opposition to a proposed data center.

Allies and adversaries

Government makes decisions about data centers at all levels. In cities and counties, zoning fights draw a disparate resistance: environmentalists, folks reeling under high electricity costs, local-control advocates, and farmers concerned that data centers eat up land. They unite in opposition to projects backed by powerful corporate interests.

Some bipartisanship has extended to Congress. In mid-September, the House passed the Ratepayer Protection Act by a wide margin: 417-3. The bill requires states to consider adopting a policy that data centers pay for the upgrades necessary to meet their high electricity needs.

The three nay House votes were from progressives who felt the bill, which would establish a voluntary structure, didn’t go far enough to protect consumers. Comparable sentiment killed the Senate version, a reminder that even allies sometimes follow different strategic paths. Some proponents of tighter regulation will accept incremental progress, while others push for more sweeping change.

Politics as usual

As midterm elections approach, candidates up and down the ballot are blanketing the airwaves and digital channels with ads against data centers. President Donald Trump, for his part, has weighed in, suggesting that data center opponents want their communities to be “backwards and poor.” It’s a fine line for some Republican candidates between responding to voter concerns and not alienating the president.

If data center politics seem familiar, it may be because this isn’t the first issue to bring together such a diverse coalition: In the 1990s, opposition to the North American Free Trade Agreement did the same, uniting organized labor and environmentalists, frequent adversaries, over concerns about job losses and pollution. That alliance never extended beyond the boundaries of its singular focus on trade agreements, though. Similarly, don’t expect data centers to permanently reshape the U.S. political landscape.

Barbara A. Trish is a professor of political science in Grinnell College.

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

Former DOGE staffers say they just hacked Anthropic in less than 12 hours

3 weeks ago

A cybersecurity firm run by former members of Elon Musk’s Department of Government Efficiency (DOGE) says it has broken into internal data and messaging systems at Anthropic. Perhaps even more surprising, though, is its claim that it did so in less than 12 hours.

The startup is called Cathedral, and it classifies itself as an AI-driven cybersecurity firm with a focus on military applications. In July, Cathedral closed a $160 million funding round that valued the company at $1.4 billion, with investors including Andreessen Horowitz and Sequoia Capital.

How Cathedral got inside Anthropic

Cathedral says it was able to penetrate Anthropic by taking advantage of vulnerabilities in business software the company uses—the same software used by many other businesses. Among the data it claims to have accessed were meeting transcripts, salary information, job offers, and Slack messages discussing its AI models.

“The fact that weaknesses in ordinary enterprise software can compromise the infrastructure of even the most advanced organizations should be a wake-up call for every company in America,” a report from Cathedral read, according to The Washington Post.

Anthropic has a program inviting outside cybersecurity companies to test its systems, so this would qualify as a white-hat hacking incident. The AI giant confirmed Cathedral had accessed “a limited set” of its systems and said it has since shut down those access points. The company also said that Cathedral spent multiple days unsuccessfully trying to break into its software directly before turning its attention to third-party business tools.

Despite that, Cathedral’s apparent success in breaching Anthropic has raised eyebrows in Washington, D.C. One unnamed official told the Post: “If a bunch of kids who have just started a company can do this, then potentially, the Chinese aren’t far behind.”

Anthropic’s rocky relationship with the Trump administration

The Trump administration and Anthropic have had a rocky relationship for some time. Donald Trump labeled Anthropic a “radical left, woke company” earlier this year and ordered all federal agencies to cut ties and stop using its products. The Pentagon designated the company a “supply-chain risk,” prompting a lawsuit.

In its S-1 paperwork leading up to its IPO, Anthropic warned potential investors about the potential impact of that tension: “The political environment and our engagement with policymakers and public positions on AI regulation and safety could subject us to heightened scrutiny, political backlash, or disproportionate regulatory action.”

Cathedral’s ties to Trump

Cathedral’s founders include several people with close ties to the Trump administration. Among the former DOGE staffers running the company are Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein.

Kliger also previously served as the Pentagon’s chief data officer and was reportedly involved in the department’s skirmish with Anthropic. Elez was forced to resign from DOGE after he was linked to tweets containing phrases such as “normalize Indian hate,” “You could not pay me to marry outside of my ethnicity,” and “Just for the record, I was racist before it was cool.” (Musk rehired him despite the controversy.)

The company is reportedly developing AI software it hopes to sell to the Pentagon to bolster the military’s cybersecurity capabilities, guarding against and launching attacks on adversaries like China. Reuters also reports that Cathedral is looking to acquire a data center or to partner with one for dedicated computing power.

The midterm elections could complicate those ambitions, though. Should Democrats regain control of the House of Representatives or Senate, they could closely scrutinize contracts awarded to companies run by founders with especially close ties to Trump, potentially slowing Cathedral’s progress.

The politicians still making the case for data centers this election season

3 weeks ago

Data centers might be the boogeyman of the 2026 election. As anxiety about the impact of AI on the economy—and of hyperscaler facilities on local communities—ramps up, politicians on both sides of the aisle have distanced themselves from the industry.

But not everyone. Even as Americans express growing discomfort with data centers, some politicians are still making the case for these massive AI projects, arguing that they represent a promising economic opportunity or are necessary to ensure America’s competitiveness. Others are taking a more conditional approach, supporting development while insisting on local approval, protections for electricity customers, or other restrictions.

The divide doesn’t necessarily follow party lines. Nor is it always a simple question of support or opposition. In Maryland, for example, Republican gubernatorial candidate Dan Cox is calling for a moratorium on hyperscale projects, while Democratic governor Wes Moore has established new requirements for developers, insisting that they “pay their own way.” Both candidates have previously supported measures encouraging the industry.

Fast Company identified a range of politicians who have championed data centers, either recently or during the current campaign. Some remain enthusiastic supporters of their expansion, while others are emphasizing stricter conditions as the backlash grows and the midterms draw closer. Still others have shifted their approach considerably. We reached out to every candidate’s office for comment.

Bruce Blakeman, Republican candidate for New York governor

Bruce Blakeman, who is challenging New York governor Kathy Hochul, might be one of the most vocal champions of data centers this campaign season. Blakeman, an attorney, argues that data centers represent a major economic opportunity for the state. “What people don’t understand is they hear that it requires a lot of energy. It does, but many of these developers are willing to put their own generation into those plants,” he told one outlet.

Blakeman has criticized Hochul’s executive order establishing a moratorium on permits for new hyperscale data centers, and has pledged to reverse it. At the same time, he’s said local communities should have a say in whether projects move forward. His enthusiasm for the industry has become a target of Hochul’s campaign attacks. “He wants fossil fuel generating factories spewing smoke everywhere,” Hochul said during a recent debate, “and he wants data centers using our rare farmland.”

Tommy Tuberville, Republican candidate for Alabama governor

Tommy Tuberville, a former football coach and current U.S. senator running for Alabama governor, has faced criticism for his evolving position on data centers.

At times, he’s suggested that, as a pro-business candidate, he’ll clear the way for data center construction throughout the state. “We’re looking at several other things we can do to bring money in; everybody brings up these data centers,” local media reported the candidate saying. “It’s one of the few ways that we can make millions of dollars for this state.” Tuberville has even suggested that China may be amping up anti-data center sentiment.

At the same time, Tuberville has proposed limits on the industry, including protections for electricity customers. He’s also argued that developers should pay for the infrastructure they require and that local communities should have a say in new projects. His Democratic opponent, Doug Jones, is advocating for a data center moratorium and has criticized Tuberville over stock trades involving companies with ties to the AI and data center industries. Financial disclosures document some of the investments at issue, although Tuberville has since divested from at least one of the companies.

Rick Jackson, Republican candidate for Georgia governor

Rick Jackson, a businessman running for Georgia governor, supports data center development but has emphasized local control over construction.

“If your community doesn’t want a data center, I’ll defend you from doing it. If you want one that has economic benefit to you, I’ll negotiate it personally to make sure you get the best bill,” he said just before a campaign event. He’s also suggested that there is a lot of “misinformation” about the negative aspects of data centers.

Notably, Jackson has faced questions about his investment in a data center project in Texas that’s drawn backlash. The Atlanta Journal-Constitution reports that another AI deal, which Democrats have pounced on, includes a 15 percent stake that could be worth $75 million. The billionaire has defended the project as a passive investment in a project with the potential to raise property tax revenues in the area. Still, he’s running in a swing state, and his AI industry ties may prove influential in the outcome.

Treg Taylor, Republican candidate for Alaska governor

Treg Taylor, a former Alaska attorney general, has publicly called for embracing data centers, albeit with conditions. He’s argued that these facilities, as long as they generate their own energy, could potentially sell power to other customers while also creating jobs. “It doesn’t have to drive up costs if you do it correctly. It can produce a net savings to the ratepayers in Alaska,” Taylor has said, according to comments transcribed by the Alaska Beacon.

Dave Bronson and Josh Church, Republican candidates for Alaska governor and lieutenant governor

Dave Bronson, a former Anchorage mayor running for the same seat as Taylor, has also spoken favorably about data center development. While acknowledging potential downsides, including noise, electricity demand, and cooling challenges, he’s argued that there’s “a lot of money” in building these facilities, according to the Alaska Beacon. He’s also emphasized the importance of independent power generation and addressing environmental concerns.

Josh Church, who is running alongside Bronson as a candidate for lieutenant governor, tells Fast Company that data centers provide a national security benefit and help ensure the U.S. isn’t dependent on foreign technology. Like other campaigns throughout the country, the team argues that building data centers could help lower nearby energy rates.

“Alaska has historically struggled to attract manufacturing and other industries because of high energy costs and expensive shipping routes to market. Digital industries change that equation,” Church tells Fast Company. “With high-speed fiber optics and potentially satellite communications, Alaska can export digital products and services worldwide without the expense of physically transporting goods thousands of miles.”

Josh Shapiro, Democratic candidate for Pennsylvania governor

Governor Josh Shapiro, running for reelection in a swing state, has in recent months taken a tougher approach to data center development. In August, he signed an executive order imposing stricter requirements on proposed projects, including energy standards and local approval. The order also removed AI data centers from a state program designed to accelerate permitting.

“Governor Shapiro has taken bold action by enacting the strictest guardrails in the nation to hold data center developers accountable and protect Pennsylvania communities,” campaign spokesperson Manuel Bonder tells Fast Company in a statement. “The Governor has made clear he will not allow developers to bulldoze our Commonwealth, bully our communities, or drive up our utility bills – and he will continue to use his authority to protect Pennsylvanians.”

Still, this is a notable reversal for Shapiro, who once championed these projects. For example, he previously lauded and approved accelerated reviews for a $20 billion effort by Amazon to build data centers across the state, according to Spotlight PA. 

Jay Obernolte, Republican candidate for U.S. House, California

Rep. Jay Obernolte, running for reelection in California, has lauded data centers and suggested that they’re important for ensuring America remains the “global leader in AI.” His proposals have focused on the practical challenges of expanding data center infrastructure, including locating facilities closer to power plants and assessing the potential for liquid cooling, which could help address the heat produced by AI facilities.

Marsha Blackburn, Republican candidate for Tennessee governor

Marsha Blackburn, a longtime Tennessee senator now running for governor, has been an explicit supporter of data centers, calling for expanding nuclear power to support the industry’s growth. She’s also advocated for greater use of coal power more broadly. “Data centers are going to be a good and important part of Tennessee’s economic growth, but we’ve got to be thoughtful about their placement,” she said in a campaign video released earlier this summer.

That doesn’t mean the Republican senator wants them built everywhere. She’s fought plans to build a facility near the Nashville Zoo, and she recently told The Tennessean that data center development must balance innovation with responsible growth and “always serve the interests of Tennesseans first and foremost.”

Americans are being cyberscammed out of billions and few are getting their money back

3 weeks ago

The U.S. government isn’t doing enough to fight the surge in cyberscams, and partnerships with the private sector remain fragmented and voluntary, an investigation into the global scam industry led by the Joint Economic Committee of the U.S. Congress has found.

The government doesn’t even have a unified accounting of how much money has been seized from overseas scam compounds, nor how much has been returned to victims, according to the report, released Tuesday by ranking Democrats Sen. Maggie Hassan and Rep. Don Beyer.

Citing findings from a yearlong investigation into global scams by The Associated Press and FRONTLINE, the report called for greater collaboration across agencies, support for victims, and deeper private sector engagement.

“Transnational criminal organizations run massive compounds that exploit U.S. technology, including AI, to carry out scams against Americans with devastating consequences,” Hassan said in an email. “The scale of this exponentially growing challenge requires Congress, the Administration, and the private sector to all strengthen their efforts.”

The report, which the AP obtained before its release, made clear how hard it is for scam victims to get their money back, despite the quickening pace of seizures announced by a variety of U.S. government agencies.

The FBI told the committee that it returned $336 million to scam victims in 2025.
The Federal Trade Commission reported that it had returned $1.83 billion to fraud victims in fiscal year 2025, but couldn’t determine how much of that money had been lost to overseas scam compounds.

The U.S. Secret Service returned more than $770 million in seized assets to fraud victims in fiscal years 2024 and 2025, but noted the money had come from a range of fraud schemes.

The U.S. State Department said that foreign law enforcement agencies that receive support from the department had seized proceeds “in the low hundreds of millions of dollars,” but told Congress to ask the Departments of Justice, Homeland Security and Treasury for details.

Taken together, that only adds up to a fraction of what victims have lost. Americans reported cyberfraud losses of nearly $18 billion — that’s more than $2 million an hour — last year to the FBI. Many don’t even report the crime.

The report also expressed concern that even as criminal scam networks expand into more countries, U.S. State Department cutbacks eliminated at least one anti-scam program in Bangladesh, which had been training local law enforcement to conduct cyberfraud investigations.

Bangladesh is an emerging hub for global scam operations, according to a report last month by the United Nations Office on Drugs and Crime that was funded by the European Union.

The U.S. State Department, which said it has 60 employees in the U.S. and Asia who dedicate significant time to fighting overseas scam compounds, told the committee that it “has the tools it needs to combat the threat internationally.”
The report urged Congress to pass pending legislation that would mandate a unified federal strategy to combat scams and establish a central location for victims to report losses and seek help. The White House has also advocated for greater federal coordination.

Several Justice Department officials told Congress that private sector companies should focus more on scam prevention, but U.S. regulators and lawmakers have so far not created clear incentives for them to do so. Unlike the 27-nation EU, the United Kingdom, Australia and Singapore, the United States has stopped short of holding technology, telecommunications or financial services companies potentially liable if they fail to do enough to protect consumers from scams.
Kate Griffin, a director at the Aspen Institute’s financial security program, said that there has been important progress, but the U.S. still lacks “durable, sustainable scam prevention.”

“We don’t see that showing up yet in reduced victimization,” she said. “And that’s what really needs to happen.”

This story is part of an ongoing collaboration between The Associated Press and FRONTLINE (PBS) that includes the documentary “Scammed.” Watch the documentary at pbs.org/frontline, in the PBS App and on FRONTLINE’s YouTube channel.

—Erika Kinetz, Associated Press

Anthropic’s IPO strategy? Cloak itself in altruism

3 weeks ago

Anthropic’s cofounders, including CEO Dario Amodei, left OpenAI because they believed AI safety features should be developed alongside model capabilities from the very start of the development process. This mission has shaped the company, and has carried through to the way it’s framing its IPO, which could happen before Thanksgiving. 

In its S-1 filing, Anthropic tells the story of a company that wants to benefit all of the world’s people, but can do so only if it goes big and leads the market. That smacks of effective altruism, a twenty-first-century philosophical approach that’s caught on with young Silicon Valley founders. Several Anthropic researchers told Wired in June that it’s important Anthropic stay in the lead among frontier artificial intelligence labs so that mass market AI will be built safely. Anthropic’s leaders believe that if the lab can reach major research milestones first, they can authoritatively define the safety regime needed to guard against the risk of the new capabilities.

But developing such safety regimes costs time and money. Challenged to maintain its lead, Anthropic might be tempted to make compromises on safety. Releasing a model quickly may improve Anthropic’s competitive position among AI labs but leave less time for safety testing. Or, restricting a model’s capabilities could reduce safety risks but limit its commercial value. 

Competitive pressure could encourage faster model development even when safety questions remain unresolved. The company has already shown a vulnerability to such pressure. In Anthropic’s original Responsible Scaling Policy, the company pledged that it would not develop or deploy certain advanced systems if it could not implement adequate safety measures. But in February 2026 Anthropic removed that categorical commitment from its policy, explaining that it couldn’t commit to halting development because it couldn’t count on its competitors doing the same.

Becoming a public company could itself complicate Anthropic’s safety ideals. Wall Street will demand constant quarterly revenue growth, not just to bolster investor confidence but also to offset the massive computing and infrastructure costs. The best way to do that is to hold onto existing customers and attract new ones, and that sets up another incentive to release ever more capable models as quickly as possible. 

That’s apparently what Anthropic has been doing. The S-1, which registers its securities with the Securities and Exchange Commission, shows that top-line revenues grew from approximately $4.6 billion for 2025 to $16.3 billion in just the first six months of 2026. 

Eventually, profits will matter just as much as revenue. Anthropic will have to develop and serve models in a way that generates bottom-line profit. The company reports a net loss of $80.7 billion for the first half of 2026, but says that only $693 million of that was lost from actual operations. The vast majority of it —more than $72 billion—came from shifting valuations of the complicated financial instruments (convertible notes, derivatives, etc.) that Anthropic has been using to raise money for future computing power and infrastructure build-out. The S-1 says that increases in Anthropic’s valuation caused increases in the fair value of its convertible promissory notes. 

At any rate, Anthropic’s proposition to investors depends on the demand for AI continuing, and accelerating. Today, there is no shortage of demand from the enterprise customers Anthropic wants to serve, but just because big corporations continue to buy access to more models and deploy AI in more ways doesn’t mean that the pace of adoption will continue. That depends on how quickly workers can adapt to new ways of doing business and gain trust in the accuracy of the AI.

And AI may not be delivering the broad productivity increases the industry is promising. Just two days before the Anthropic IPO, Apollo Global Management dropped a report saying that “despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.” The firm’s chief economist, Torsten Slok, adds that revenue per worker is rising for the big seven tech companies, but is flat for the “S&P 493.” 

On the other hand, Apollo may be missing the point. What’s important for Anthropic is that enterprises continue to build their AI stack even if the returns aren’t yet hitting its balance sheet. Many of them are. According to the S-1, Anthropic has more than 11,900 customers—including half of the Fortune 500—that paid $100,000 or more over the past 12 months, with more than 1,900 of them paying $1 million or more over the same period, and approximately 250 of them paying $10 million or more.

Biggest stock listing ever? A look at Anthropic’s IPO by the numbers

3 weeks ago

Anthropic has the chance to make history with what could be the biggest initial public offering of stock ever.

The artificial intelligence lab filed papers Monday setting up a Wall Street debut next month, ahead of its chief rival OpenAI and just months after SpaceX set records in June with an initial public offering that raised $86 billion.

In a 335-page prospectus, the company behind the Claude chatbot outlined rapid revenue growth, but an inability to turn a profit yet. The offering could bring in as much as $100 billion, according to reports, which would help to accelerate the expansion of the five-year-old startup.

The filing comes as Anthropic’s competitors like OpenAI are holding off on going public until next year, and as fears rise that the technology could slip free of human control to carry out harmful actions. Anthropic CEO Dario Amodei said the industry should slow its fast-moving development to give safety measures time to catch up.

But the company’s massive spending and revenue point to strong momentum.

Here’s a look at the key figures behind the lab’s ambitions.

Up to $2 trillion

That’s the reported valuation the company could reach once it goes public. This would surpass SpaceX’s initial valuation of $1.77 trillion and make Anthropic one of the world’s most valuable public companies. Nvidia, a maker of computer chips, leads that pack with a current valuation of more than $5 trillion.

Anthropic’s final valuation and offering price have not been established.

$131.7 billion

Anthropic said in the filing it had this amount available as cash or cash-equivalents as of June 30. The company also said its plans to build out compute and infrastructure are “changing the cash flow profile of our business” and will mean high capital expenditures.

$16.3 billion

This is Anthropic’s revenue in the first half of the year, which is about 14 times its revenue from the same period last year.

$80.7 billion

The amount of money Anthropic reported in net losses in the first half of the year. This included a one-time, $72.4 billion charge for an accounting adjustment related to financing agreements, according to the filing. It’s not uncommon for tech companies going public to not yet have turned a profit.

93

This is how many pages Anthropic used to enumerate risk factors for investors. Public companies are required by the Securities and Exchange Commission to disclose risk factors in their prospectus, but the lengthy nature of those in Anthropic’s is unusual.

Among the concerns were that advanced AI systems could pose “catastrophic or existential risks to humanity.”

250

The approximate number of customers Anthropic has that each account for $10 million or more of revenue in the past year. Anthropic did not disclose the names of its biggest customers in the filing, but it pointed to its top five customers representing approximately 25% of its revenue this year.

50.1%

The voting power Amodei and six other co-founders of Anthropic will hold at the company after it goes public. The company will issue a single share of a Class F common stock to a limited liability company organized by its co-founders. Amodei’s sister and Anthropic’s president, Daniela Amodei, is among the former OpenAI employees who left the company to create Anthropic in 2021.

—Associated Press

Why is Whoop pivoting to be doctor’s helper? Because that’s where the money is

3 weeks 1 day ago

When former Harvard squash cocaptain Will Ahmed developed the idea for wearable company Whoop in the early 2010s, he was focused on helping serious athletes like himself optimize their performance. The wristband he designed tracks activity, sleep, and recovery. But in August, the Whoop founder and CEO turned his attention to a very different audience: Medicare recipients.

The Whoop Connected Care offering—which includes a Whoop membership, a Whoop device, and a third-party blood pressure monitor and cuff—is now a pilot program available to Medicare beneficiaries 65 and older who have been diagnosed with hypertension, prediabetes, high cholesterol, or obesity. “Continuous data matters,” says Dr. Ami Bhatt, a board-certified cardiologist who joined Whoop as chief medical officer over the summer. “People want to perform and age well, and that’s what we’re aiming for.”

Bhatt acknowledges that at first glance, American seniors appear to have little in common with the athlete ambassadors—like international soccer star Cristiano Ronaldo and Super Bowl-winning NFL quarterback Patrick Mahomes—most closely associated with the Whoop brand. But women older than 60 are one of Whoop’s fastest-growing customer segments. They’re bellwethers of a broader shift as consumers’ pursuit of wellness morphs into a deeper interest in preventive healthcare. Wearable companies, looking to cash in on the trend, are hiring doctors like Bhatt and courting clinical partners.

Whoop and Oura, among the buzziest wearables on the market, are already multibillion-dollar businesses. Whoop was valued at $10.1 billion in March, when it raised $575 million in Series G funding. Oura was valued at $11 billion at the time of its $900 million Series E round last year.

But there are only so many weekend warriors and wellness hobbyists willing to pay hundreds of dollars a year to track their steps and minutes of deep sleep. The key to attracting bigger audiences and keeping them engaged, wearable companies posit, is to connect the devices’ data to meaningful medical insights. And if that also allows a company like Whoop to tap into Medicare’s $1.2 trillion budget, all the better.

Bhatt, who previously served as inaugural chair of the Food and Drug Administration’s Digital Health Advisory Committee and chief innovation officer at the American College of Cardiology, says wearable companies shouldn’t position themselves to replace clinical care. But they can certainly supplement it. “What we need to be doing is helping people understand their baseline and helping clinicians understand what’s happening in between visits,” she says. “I want them to get a comprehensive addition and context to the story since they last saw that patient.” 

Silicon Valley is home to a graveyard of abandoned hardware devices, left behind as consumer tastes change. Whoop and Oura plan to escape that fate by embedding themselves more deeply in healthcare. 

Not-quite medical devices

The backdrop to this expanding mandate for wearables is a simmering debate over regulation. Historically, the FDA has clearly delineated between general wellness devices and medical devices. Medical devices, which are subject to stringent rules and approvals, are designed to diagnose and treat specific diseases and conditions. Wellness devices, in contrast, are designed to support healthy lifestyles more broadly. Wearables like Whoop and Oura are deliberately upending that distinction. 

“The latest generation of wearables delivers clinical-grade insights,” Oura CEO Tom Hale wrote in a Wall Street Journal op-ed last December calling for a change in federal policy. 

Whoop, for its part, stood its ground when the FDA sent a warning letter last summer questioning a feature the company had built to track blood pressure. The company edited its app interface to clarify that the feature was not diagnostic, but otherwise refused to compromise. 

In January, the FDA capitulated. It issued new guidance that gives wellness devices far greater leeway to measure the physiology associated with chronic conditions. Six months later, the agency dropped its enforcement action against Whoop’s blood pressure feature. 

[Image: Whoop]

The companies have been racing to take advantage of the friendlier rules and solidify their positioning as healthcare solutions. In May, Oura unveiled a partnership with virtual care provider Counsel Health. When a user’s biometrics show what Oura describes as “major signs of strain,” based on measures like body temperature and resting heart rate, the device maker now sends an email suggesting that it’s time to “Connect with a Doctor who has context.” 

As for Whoop, members can upload their medical records; complete blood work with partner Quest Diagnostics to evaluate more than 100 biomarkers; and connect with doctors on-demand. Whoop’s top-tier device is named the Whoop MG, short for “medical grade”—even though it’s not an FDA-approved medical device.

Data meets reality

Researchers and clinicians caution that a gap remains between the products as they exist today and the practical realities of delivering care. 

Sandeep Kishore, a physician-scientist and associate professor at the University of California, San Francisco, is part of an effort to reimagine care for chronic conditions with the help of digital tools. Wearables have “potential,” he says, but at the moment we’re in what he describes as a “twilight zone.” Devices are generating reams of data, but the algorithms behind that data are typically proprietary and difficult to independently validate. 

For example, Whoop offers a feature called Blood Pressure Insights, and Oura offers what it calls NighttimeBP. Because the companies say they have designed these data insights for wellness purposes, and not for the purposes of clinical diagnosis, the features are exempt from the rules that govern traditional blood pressure cuffs. 

“Nighttime BP does not display mmHg values,” Oura explains in its FAQs, referencing millimeters of mercury, the standard unit of measurement for blood pressure. “It shows dipping patterns and trends, not absolute blood pressure measurements.”

But that isn’t stopping patients—er, wellness consumers—from bringing these measurements and others to their doctors. “I’m having a bunch of unvalidated data coming in to me, and I’m being asked to react to that,” Kishore says. 

Emerging research suggests that even more established wearables still suffer from accuracy problems. A paper published in Nature this past January found that Apple Watch excels in some physiological and behavioral measurements, like heart rate at rest, but fails in others, like energy expenditure. 

Chronic conditions and more

Despite such shortcomings, device makers are finding that some healthcare providers are willing to experiment. After all, three in four American adults lives with a chronic condition. “The current system isn’t working,” Dr. Chris Curry, a board-certified ob-gyn and clinical director of women’s health at Oura, says of the refrain that she hears from Oura’s clinical partners. “Let’s try something new.” 

Oura has built a feature for users taking GLP-1 medications, for example. “We’re not trying to replace the doctor prescribing the GLP-1,” she says. “But the physicians who are in that world have a hard time keeping track of all their patients. And so we built the GLP-1 experience to be part of the solution. ‘My body’s changing really quickly—is everything okay?’”

Curry can imagine Oura playing a deeper role in the future. “The literature suggests that not every med is perfect for every person,” she says. “We don’t know why yet. I think [Oura] is well positioned to ask questions, and then ideally find the answers. Are there optimal dosings for certain categories of people? Are there things that are happening under the hood to your cardiovascular system or your metabolic system that are differentiated from weight loss, and we should also be tracking or managing in different ways? We want to be part of the scientific literature that lets physicians more smartly prescribe.”

Bhatt, meanwhile, has been seeking collaborators in specialties she sees as priorities for Whoop, including cardiometabolic health, women’s health, and neurology. The company is also teaming up with the National Council on Aging and AARP to tailor its user experience for older audiences.

“They say 80 is the new 50, right?” Bhatt says. Or it could be, if Whoop can prove that its devices can improve health outcomes. Deliver on that promise, and everyone—consumers, doctors, insurers—will be knocking on her door.

How much will data centers increase your electric bill?

3 weeks 1 day ago

One of the biggest reasons most Americans don’t want a data center built nearby is that they worry it will drive up their utility bills. But whether those bills rise, and by how much, depends on the local power grid and the terms of the deal.

In fact, the impacts have at times been counterintuitive, with new data centers in some cases actually driving down utility bills for other customers on the same grid.

Consider North Dakota. A 2025 study from Lawrence Berkeley National Laboratory and the Brattle Group found that increased electricity demand from large customers, including data centers, helped lower bills in the state between 2019 and 2024. That’s partly because North Dakota had a relatively abundant supply of electricity, including wind power, and didn’t necessarily need to build new infrastructure to accommodate the additional demand. Instead, utilities could spread their existing costs across more customers, bringing bills down. (That dynamic becomes less likely as the grid runs out of spare capacity, but more on that later.)

The study found that over that five-year period, there were several reasons that prices were rising other locations. Inflation was one obvious driver. But there were a range of other significant causes, from Russia’s war in Ukraine impacting the price of gas for power plants to the cost of utilities making the grid more resilient to wildfires in places like California.

To get a clearer picture of how data centers affect electricity costs, researchers at the University of Pennsylvania’s Wharton School and Harvard Business School took a different approach. In a new working paper, they compared electricity bills in places where data centers were built between 2020 and 2025 with bills in areas that didn’t see new construction. “We started looking at this earlier this year…there was a lot of public perception about what data centers are doing, but strong academic and critical evidence on what the real impacts are was missing,” says Vishrut Rana, a Wharton doctoral student and one of the study’s authors.

That research has found that new data centers had a relatively small impact on residential bills over the study period—on average, just 4% of the total observed increase. (Industrial energy prices jumped up higher.) The impact varied depending on location and the type of data center. Colocation data centers, where businesses rent space to house servers, had more impact on prices than hyperscale data centers for single tenants. That may be both because hyperscalers have tended to build in more rural areas where there was more room on the grid, and because hyperscalers have also been more likely to sign contracts where they cover the cost of adding new infrastructure—sometimes new wind and solar power that can replace more expensive gas generation.

The risks of a grid stretched to its limits

That doesn’t mean the effect on energy bills will be small going forward. Another new report from a nonprofit called ESIG, the Energy Systems Integration Group, looks at multiple factors that influence what could happen because of a hypothetical new data center. First, in theory, there still might be some places where there’s enough extra power infrastructure on the grid that a data center might lower costs. The report’s modeling suggests that scenario could lower rates by 0.88 cents per kilowatt hour in the example of a hypothetical utility in the Southeast. But because of the scale of demand for electricity, if those places exist, they probably won’t for long. And a grid might need that extra capacity for other new uses, too.

Several factors could push bills higher. A data center might overestimate how much electricity it needs, so new energy infrastructure gets built and other ratepayers get stuck helping cover the bill. A data center might be built late or shut down earlier than expected, again leaving other ratepayers covering the cost of new infrastructure. In some regions, utilities don’t own all of their own power plants and have to buy energy in auctions; as data centers sharply increase demand, wholesale prices go up, and that gets passed on to customers.

The type of power plants that are being built to support data centers can also affect costs for consumers. Energy Innovation, an organization focused on clean energy, has estimated that relying on fossil fuels for new energy generation would add $29.7 billion a year to customer bills by 2030; clean energy could save around $5 billion a year. If coal and gas prices spike, as they have in recent years, the cost could jump to $40.5 billion a year. In that case, clean energy could save as much as $13.5 billion.

In a worst-case scenario where multiple factors are simultaneously pushing electric rates higher, the ESIG report estimates that it could raise rates by 1.32 cents per kilowatt-hour. (It doesn’t translate that into monthly bills, but that might mean paying roughly $10 more a month.)

[Image: ESIG] Who pays for all that power?

Other factors could mitigate rate increases. States and utilities are beginning to require contracts with minimum billing terms, so data centers wouldn’t end up paying less if they happen to use less energy. Other contracts include early termination fees. Data centers or their customers can also cover the cost of new power plants. If federal regulations change, it could also be easier to charge data centers for upgrades to transmission networks.

The terms of these contracts are evolving rapidly. “What’s happening is actually quite remarkable [in terms of] how quickly things are changing,” says Trieu Mai, senior director of programs and research at ESIG. More than a dozen states, from Tennessee to Minnesota, have passed laws designed to protect consumers from rising energy bills. Hundreds of data center bills were filed this year. In some states, attorneys general are also challenging data center contracts that they say don’t adequately protect consumers.

Utilities are also changing their approach, though the details differ everywhere. In North Carolina, where Duke Energy has said that data centers are responsible for more than 85% of load growth on the local grid, the utility initially planned to raise rates 18% next year (the cost also included upgrades like making infrastructure more resilient in storms). Under pressure from the governor, that was negotiated down to around 9.5%. Now, the utility has proposed a new tariff for large customers like data centers, though it won’t apply to contracts signed before July 2026, and critics say that it doesn’t go far enough—it doesn’t cover the cost of new transmission upgrades, for example, or what might happen if a data center leaves early.

These changes represent a major shift in how utilities typically operate, and they may not happen everywhere. “Utilities are spending billions of dollars on new infrastructure for data centers, and the basic utility business model is to spread the cost of new infrastructure to everyone,” says Ari Peskoe, director of the Electricity Law Initiative at the Harvard Law School Environmental and Energy Law Program.

Even when companies claim they’ll cover the cost of new grid infrastructure, it’s difficult to verify whether that’s actually happening. “It’s very hard because a lot of the details are confidential,” says Peskoe. “And then there’s not a lot of transparency about what’s actually being built for data centers, particularly when it comes to the high-capacity transmission network, interstate transmission system that connects everyone together. Utilities are not really doing a good job of disclosing what is being built for what data center. The federal regulator is ordering more transparency, but that’s still a process. So TBD on how that all plays out.”

The bottom line: Transparency is critical if residents want to understand what a proposed data center could mean for their utility bills. And at a time when data centers have become a defining issue in the midterm elections—and most Americans are convinced that data centers will significantly spike bills—there’s no simple answer to what impact a large data center might actually have. “The answer would be: it depends,” says Mai. “Of course, it’s an unsatisfying answer. That’s why you have to do the math to see what it depends on for your region.”

AI agents have no rules, so Meta wrote its own

3 weeks 1 day ago

Why would anyone trust Meta?

After all, this is the company that became the world’s avatar for shameless and exploitative data harvesting. The same data mastery that powers one of the most effective advertising platforms ever made also gave us Cambridge Analytica, whistleblowers like Frances Haugen and Sarah Wynn-Williams, and unflattering films like The Social Dilemma and, just this month, The Social Reckoning. In the Axios-Harris Poll 100, Meta scored a lowly 59.6, down in the basement with ByteDance and X.

Publishers have their own reasons to be suspicious. Meta once vastly increased the reach of its properties, and even paid them for content, then abruptly took it all away when its priorities shifted. Meta has blocked news entirely in Canada since 2023, and briefly did the same in Australia in 2021.

So when Nieman Lab reported that Meta had changed its terms of service in a way that seemed to give itself permission to scrape the site of any company with a Meta Business account, nobody was inclined to give it the benefit of the doubt.

The new terms, which allow Meta to use automated tools to “access, index, cache, analyze, collect, retrieve, or crawl” websites “associated with” a business account, and to use that content to “provide, develop, improve, research, train” Meta products, could be interpreted as signing over to Meta the license to do whatever it wants with your content, including training its AI models or even offering up AI answers based on it. Indeed, three legal experts Nieman Lab consulted warned publishers of exactly that danger.

{"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":""}}

Meta quickly backpedaled, sort of. Within a day, it removed the word train and limited scraped content to its advertising and business products. A spokesperson insisted, “This is not about scraping news sites.” That didn’t calm anyone down. Legal experts said the revised language still permits mass scraping, and publisher groups on both sides of the Atlantic demanded a pause and filed complaints with regulators. This week Meta pushed the effective date from October 30 to November 20. The U.K.’s News Media Association called the delay “a pause, not a solution.”

What Meta probably wants

It’s unlikely that Meta secretly changed its position on what has been, for the past three years, the single most important legal issue in how AI affects media, and then tried to bury it in a terms-of-service update. The impetus for the change was likely more mundane: Meta says the crawling language came from folding its Ad Creative Generative AI Terms into its main commercial terms, and AI tools that build ads for a business presumably need to read that business’s website. As Meta pushes further into agents like its Muse assistant, its bots will need to touch the open web more, not less.

But you don’t have to be a conspiracy theorist to see how Meta stepped in it here. The change isn’t irrelevant, and publishers should be concerned about how legal agreements are worded. And some of Meta’s wording is hard to explain away. The terms let Meta create “test credentials,” which Cornell law professor James Grimmelmann said would let it sign up as a fake user to get past registration walls.

There’s no opt-out short of abandoning Meta’s business tools. And the change arrived just as Meta began charging some accounts, through its new Meta One subscription, to post links to the same content its bots could crawl for free. And, as I said, Meta has wholly earned its reputation for not considering all the dimensions of how its actions can lead to disastrous consequences.

What Meta did here, and the media’s understandable reaction, is a symptom of a deeper problem. Some will blame AI companies’ thirst for data, or a Machiavellian quest to squeeze the media industry, but the real issue is a vacuum of standards and rules around what AI agents, and the crawlers that power them, can and cannot do.

Whoever holds the leverage writes the law

You can see this same issue in another freakout from the past week: an Axios report that banks were seeing a growing number of “unauthorized” AI agents accessing customer accounts. While that sounds scary out of context, what it mostly represents is individuals using AI services to access their accounts and take actions on their behalf, not some kind of mass hacking event.

Still, there are real concerns around agents the bank didn’t sanction accessing customer accounts, not the least of which is liability when something (inevitably) goes wrong. That’s why JPMorgan Chase CEO Jamie Dimon told analysts earlier this month that letting in outside agents is “just too risky for clients and for JPMorgan,” and the bank now blocks them from logging in. Schwab is doing the same while steering clients to its own AI assistant (a move that also happens to protect its lucrative cash business), and Citi doesn’t allow third-party agents to log in either. One of the agents getting locked out, incidentally, is Meta’s own Muse.

In both cases, the platform saw either an opportunity or a gap with respect to AI usage, and as the party with leverage, it decided the rules. Which is exactly what happens when there are no rules—the person with the biggest gun gets to make them. When it comes to AI, and especially agents, we’re all living in Tombstone.

We saw the same dynamic in last year’s dustup between Perplexity and Cloudflare. To recap, Cloudflare accused Perplexity of using stealth crawlers to get around sites’ bot-blocking preferences, and Perplexity insisted its agents were legitimate proxies for actual users, entitled to the same access.

Regardless of who you think is right, it’s clear Perplexity had leverage when it was them versus individual sites. But when Cloudflare got involved, the equation changed. As the gatekeeper for a huge share of the web, Cloudflare delisted Perplexity as a verified bot and gave every customer a way to block its stealth crawlers. Perplexity has since deputized itself in another matter, blocking ads that Time served to AI agents as “deceptive” without saying how it defines the term.

Who gets to be sheriff?

Certainly, the system would benefit from common rules and standards, and—to stretch the Wild West analogy—we’re beginning to see the townspeople band together. Collective organizations like SPUR (the Standards for Publisher Usage Rights coalition), now more than 80 publishers strong, are looking to establish standards for how publisher content is tracked and valued when AI systems use it.

And lawmakers are moving, too: The bipartisan Stealth Bot Prohibition Act in Congress would require crawlers to identify themselves and their purpose. And several states, including New York, have already passed their own versions, so any site that crawlers scrape—not just publishers—will have a better idea of what to do with them.

What this means for Meta and anyone else attempting to use their leverage to set the rules: You overreach at your peril. While legal departments will always default to grabbing the widest possible set of rights, that will arouse suspicion and mistrust in a world with no rules to govern those rights, and it’s why it’s not enough for a Meta spokesperson to simply say publishers can still block its crawler with robots.txt. Putting those words in the actual terms of service, as WAN-IFRA, the global publishers’ association, has demanded, would make clear to everyone what rules you think you’re playing under.

The fact that those rules are up to Meta at all is the real vacuum. Right now, whoever holds the leverage decides what an agent can do with what it collects: Meta through its terms, JPMorgan through its login page, Perplexity through its index. A three-week delay doesn’t change that. Until someone builds a framework everyone has to follow, you can’t blame publishers for circling the wagons whenever someone pins on a badge and declares themselves the new sheriff in town.

{"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":""}}

NBA partners with Kalshi and Polymarket as official prediction markets

3 weeks 1 day ago

The NBA has struck deals that will make Kalshi and Polymarket the league’s preferred prediction markets, along with an agreement with a federal agency tasked with regulating those exchanges.

Those agreements, announced Monday, are intended to provide “significant integrity protections for NBA-related markets,” including limitations on what can be offered on those exchanges, transparency when it comes to suspicious patterns by site users, and banning league and team personnel from making trades related to league events.

The prediction markets will have “information-sharing obligations relating to investigations and suspicious trading,” the NBA said. The NBA has had similar guardrails in place with its sportsbook partners for some time and has cited instances where that system has flagged irregularities in betting patterns.

“If you’re going to be betting on our games or trading on our games or playing fantasy around our games, it’s better for it to be regulated than unregulated,” said Scott Kaufman-Ross, the NBA’s head of media distribution and partnerships. “And clearly, our fans have interest in this activity. So, we’re in a better position to protect the integrity of our games in a regulated market where there’s transparency, where things can be monitored and we have insight into what’s happening.”

The NBA is also expanding its agreements with current official sports betting partners, such as DraftKings, to include their prediction-market offerings. Not all states offer legalized sports betting but predictions markets—at least on some level—are essentially available in all 50 states.

More than a dozen states are involved in litigation over prediction markets, and earlier this month the NFL asked the Supreme Court to rule on the regulation of such markets. The U.S. Commodity Futures Trading Commission contends that it has exclusive jurisdiction on oversight, though some states believe it should fall under their purview.

Earlier this month, the NBA told sportsbooks they will not be permitted to offer certain types of wagering this season, including but not limited to which players record the first stat of a game, the outcome of the opening tip, and taking the “under” involving players not expected to play more than a few minutes. The deals announced on Monday are expected to have similar restraints.

“A big part of this is a limitation on the types of markets that can be offered,” Kaufman-Ross said. “We will have the right to approve the types of markets that are available on our games, which is a very important component for us. And we’ve had very constructive dialogues with the companies as part of that.”

The NBA’s memorandum of understanding with the CFTC “reflects a shared commitment to ensuring that NBA-related prediction markets are subject to appropriate regulatory oversight and integrity protections,” the league said.

Kalshi and Polymarket, as part of their partnerships with the NBA, will now be able to describe themselves as the “Preferred Prediction Market Exchange of the NBA,” plus have access to league intellectual property such as logos, be able to get NBA statistical data from league partner Sportradar “to operate and settle authorized NBA-related markets,” and other marketing activations.

The terms of the NBA’s agreements with Kalshi and Polymarket were not disclosed, other than the league confirming that they are for multiple years. There are NBA stars with business relationships with those markets; Miami’s Giannis Antetokounmpo has promoted Kalshi and Philadelphia’s LeBron James recently signed a deal with Polymarket. At least one other NBA star, San Antonio’s Victor Wembanyama, recently said he will never endorse such sites.

—By Tim Reynolds, AP basketball writer

Checked
16 minutes 26 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