It’s just one of a wide range of new and advanced capabilities, OpenAI says, including mathematical proficiency, software engineering, and defensive cybersecurity. But “computer use is a particularly important part of what’s new,” OpenAI cofounder and president Greg Brockman told reporters today. The model “can zip through spreadsheets, fill out forms, and navigate across web pages often at superhuman speed,” he said.

The company shared a video with reporters of someone sitting on a chair, instructing the computer with their voice in what appeared to be a seamless, albeit staged, interaction.

Computer use “shows how far we’ve come from sort of aspirationally training for computer use to bringing real value to people every day,” Mia Glaese, vice president of research at OpenAI, said.

OpenAI also released benchmark assessments that showed Astra beats its previous best model, GPT-5.6 Sol, as well as the best current models from its rival Anthropic, Claude Fable 5.1, on a wide range of tasks. On ARC-AGI-3, a difficult benchmark that is meant to test a model’s ability to apply reasoning to situations it has not encountered before, Astra scored 98.6%, whereas GPT-5.6 Sol had scored only 7.8%, and Anthropic’s Claude Opus 5 only 30%. Astra also scored 100% on ExploitBench, a tough cybersecurity challenge, beating GPT-5.6 Sol’s 78.5% result.

OpenAI is not first to market with an AI agent that can use a computer—Anthropic pioneered the space with a beta program in 2024, and Perplexity built a system that can navigate a virtual computer to complete tasks for users in February this year—but the concept is still not a mainstream part of working on a computer for most people. Brockman said he has been thinking about the idea of an AI using a computer as a human does since he first started at OpenAI in 2015.

To Brockman, computer use could be considered the start of the “AGI era.” AGI stands for artificial general intelligence. Its exact definition is disputed, with views among AI researchers divided. OpenAI itself once defined AGI as an automated system that can perform all economically valuable work as well as or better than humans.

Brockman said that achieving AGI has not come in one big moment, as he originally predicted, but rather in bits and pieces. He noted, “It’s not unreasonable to feel that we are now in the AGI era, and I think that if you want to say this [model is] the first one, I think it’s reasonable.”

But don’t expect computer use to show up in every office overnight. OpenAI is limiting the release of Astra to a limited set of enterprise customers, including those in its cybersecurity-focused Daybreak program. Astra will become available to all Plus, Pro, and Enterprise users “in the coming days,” OpenAI says, as well as through the OpenAI API and AWS. The version available to these customers will refuse “advanced cybersecurity tasks,” according to the company.

The limited initial release will also allow OpenAI to scale its compute capacity because Astra is “a very large model,” an OpenAI spokesperson tells Fortune. Astra’s development also involved the company’s largest training run “by far,” OpenAI vice president of research Aidan Clark told reporters today. “It’s the first time we’ve pretrained on more than 100,000 GPUs at our Stargate site in Texas.”

OpenAI says it added more cybersecurity safeguards. Is it enough?

OpenAI delayed the release of Astra by an unspecified amount of time to add extra safeguards and ensure a smooth launch following July’s Hugging Face incident. Since then, OpenAI has added more monitoring and further isolated its training environments, although some say the type of monitoring it added isn’t sufficient.

OpenAI submitted Astra to the U.S. government for review ahead of release per the terms of the loosely defined AI safety framework, which is a voluntary agreement between tech companies and the Trump administration whose details have not been made public, rather than a formalized, public process.

Brockman said OpenAI did its “standard testing process” with the government, and called it “a very good partnership.” When pressed for details on what the process is, Brockman declined to elaborate. “I just don’t want to misstate anything because there’s nuances on exactly how the process works,” he said.

Astra is also OpenAI’s first model that meets its “critical cybersecurity capability threshold” under its Preparedness Framework, an internal policy that governs the safety precautions the company will put in place depending on the risks a model presents. This means Astra can find and exploit previously unknown security flaws without human oversight, under the right conditions.

For this reason, OpenAI is giving only certain partners access to Astra’s most advanced cyber capabilities, the company shared earlier this week. The version provided to the company’s Daybreak customers permits approved customers to use the model for common cyber-defense tasks. But it will not allow customers to develop exploits or take on tasks that could be used to conduct cyberattacks.

This story was originally featured on Fortune.com

“These companies can quickly integrate prediction market functionality [and] not have to spend a ton of tech resources … to build the entire infra behind it,” Jason Fung, head of Gate DexBuilder, told Fortune.

The exchange’s expansion comes amid an explosion in the popularity of prediction markets. These platforms, which allow bettors to place wagers on virtually everything, are serving as a new way to gauge public sentiment. Kalshi and Polymarket dominate the industry, and consumer-facing businesses have begun incorporating their market data and event contracts into their own products.

Gate itself first entered that ecosystem as a Polymarket partner. Earlier this year, the exchange integrated Polymarket Builder, a tool that lets exchanges and apps offer Polymarket’s existing prediction markets within their own products. Fung said Gate was the first exchange to integrate it. That experience helped shape Gate’s own offering, and inspired it to compete with its former partner. 

Besides the Event Contracts Builder, Gate has launched a $3 million grant program to support companies developing event contract products. Fung said the funding comes entirely from Gate and can help cover development costs. Selected builders may also receive technical assistance, marketing support, and media promotion. The program is still in its early stages, and Gate is reviewing its first group of applicants.

From gaming to crypto

Before crypto, Fung started his career in media and entertainment, with a focus on gaming. During that period, he lived in Canada and held roles at organizations including media and technology company BroadbandTV and e-sports website Azubu. Fung later moved to China and spent five years working for tech giants like Alibaba and ByteDance, where he worked on TikTok Global. 

He eventually entered the crypto space during the ecosystem’s 2021–22 boom and bust period, and pursued what he now regards as short-lived Web3 narratives. Exhausted by that experience, Fung transitioned to more established roles, working for the layer 1 Sei Foundation and then AP Finance, the U.S. arm of centralized exchange Bybit.

In July, Fung joined Gate to further expand the globally distributed exchange. Though remote, the 13-year-old company operates hubs in Hong Kong, Japan, Dubai, and Malta, alongside its North American entity, Gate U.S.

This story was originally featured on Fortune.com

Subsense, a four-year old company developing brain-computer interface technology, announced the news alongside Kurzweil at an event at its Palo Alto headquarters on Wednesday evening. 

“Ultimately, we want to merge [the smartphone] with the brain,” said Kurzweil, whose famous “singularity” theory posits that technology will soon allow humans to transcend the limitations of our biological brains.  

“The whole point of humans is to go to greater frontiers, expand who we are,” Kurzweil said at the event. He described external devices like smartphones as a barrier that slows down our ability to access the power of AI.

Subsense is one of numerous neurotechnology companies developing brain computing interfaces, with Elon Musk’s Neuralink most prominent in the group. Neuralink has implanted wireless chips in the brains of roughly a dozen people, enabling patients paralyzed by spinal cord injuries to control computers with their thoughts. 

While Neuralink requires drilling through the skull to implant electrodes, Subsense is developing a nonsurgical—though still quite invasive—approach that involves introducing charged nanoparticles into the brain intranasally. The process would be similar to inhaling allergy medicine via a nasal spray. Once inside the brain, the nanoparticles would be controlled—to stimulate neurons in specific areas of the brain or to read neuron activity—through a specialized cap or headset containing magnetic coils. Subsense showed off a nonfunctional mock-up of a cap that looked like a cross between a baseball cap and a bike helmet.

But the sci-fi-like technology is years away from being available, assuming it ever makes it beyond the lab. Subsense, which has raised $27 million in funding, is currently doing preclinical work, experimenting with various elements of the technology and conducting some experiments on mice. Subsense hopes to then make the technology available for medical use, and eventually make it a consumer product. 

“I believe the brain-computer interface eventually may become one of the most intimate technologies that human beings ever created,” CEO Tetiana Aleksandrova said on Wednesday. 

As a product and vision advisor at Subsense, Kurzweil will provide perspective on the company’s technology road map and on how the convergence of AI and neural interfaces will reshape human-computer interactions, the company said. Kurzweil, a National Medal of Technology and Innovation recipient, is one of the most influential thinkers in the field of artificial intelligence, with books such as The Age of Intelligent Machines and The Singularity Is Near predicting many of today’s technologies.

This story was originally featured on Fortune.com

With her trademark streaked hair parted down the middle and large aviator glasses, Steinem, a co-founder of Ms. Magazine, was a hugely recognizable presence wherever she went. And she went everywhere: Much of her life as an advocate was spent on the road, speaking at college campuses and community centers, at rallies and marches.

“You can’t do it on the phone or on the web,” she told The Associated Press in a 2015 interview. “You have to be there with all five senses. It should be obvious that people can’t empathize with each other unless we’re in the same room.”

Steinem, who had been in declining health, died Wednesday at her home in New York City, according to a post on her social media page. She had recently completed a memoir, to be released this fall.

“Gloria’s greatest gift was her ability to listen to others, to make others feel seen and heard,” the post said. “Her words, actions, and example gave people permission to be their truest selves.”

Though Steinem had recently become more frail physically, her wit and humor remained totally intact. Her friend and fellow activist, Abigail Disney, said she’d visited Steinem’s Manhattan home a bit over a week ago, as those close to her gathered by her side. “She was just as funny and just as sharp as I have ever known her,” Disney said. And, she said, generous.

“Everything Gloria did was rooted in kindness,” Disney said. “It flowed through her political activism.”

Steinem changed her career path after a start as a journalist

Steinem began her career as a journalist, notably reporting in 1963 on degrading conditions in Hugh Hefner’s Playboy empire — an undercover assignment she came to later regret. She credits a meeting six years later, where women spoke out in favor of abortion rights, with changing her career path.

“That speak-out was the beginning of activism for me,” Steinem, who had undergone a secret illegal abortion herself at age 22, told the AP in an email in 2022, reflecting on the reversal of Roe v. Wade. “Either our control over our own bodies is equal, regardless of sex or race, or we’re not living in a democracy.”

Those who watched Steinem denounce discrimination for so many years likely never suspected that she’d had to overcome a deep fear of public speaking. She would have simply stuck to writing, she said, if only editors at the time had been willing to let her write about the women’s movement.

But they weren’t interested, she told the AP in 2015. “So I ended up going out and speaking, which was my nightmare,” she said. “I mean, I was terrified. But I’m grateful, even though I still get scared even now.”

Steinem’s visibility endured throughout her life

Time hardly slowed Steinem down; through her 80s, she was traveling the globe and speaking out on issues of importance to her. These included everything from genital mutilation to Korean reconciliation, but it was equality for women that took precedence. “Being a feminist means that you see the world as whole instead of half,” she once said. “It shouldn’t need a name, and one day it won’t.”

As the years went on, Steinem’s visibility endured. A few days after her 85th birthday, she was leading a women’s “talking circle” at an off-Broadway play celebrating her life, where one by one, women in the audience — many fighting tears — rose to thank her and tell their own stories of discrimination. In 2020, her life was the subject of a feature film, “The Glorias, A Life on The Road.”

And at 91, she collaborated on a picture book with Liberian peace activist and Nobel laureate Leymah Gbowee aimed at inspiring young people to change the world. “Rise, Girl, Rise: Our Sister-Friend Journey. Together for All” was published in February.

The constant visibility came with a price; even into her older years, Steinem expressed frustration that despite her accomplishments, she was often described as beautiful or glamorous. A 1992 Vanity Fair profile even noted her “spectacular racehorse legs.”

“Nobody ever called me beautiful until I was publicly a feminist, and that was in my mid-30s,” she told AP in a 2011 interview. “It was clear that label was being assigned to me. The most hurtful part is that you work very hard, and people say it’s because of your looks.”

Steinem often responded to hostility with humor

Of course, Steinem also had to deal with hostility — sometimes from surprising places. Appearing on Larry King’s talk show on CNN in 1990, she was visibly stunned when a female caller greeted her cheerfully, then told her to “rot in hell” because she had ruined “the beautiful American family.”

But by then, she noted, at least people were taking feminists seriously. That wasn’t the case in 1972 when she co-founded Ms. Magazine and many commentators — especially male — dismissed it as ridiculous, predicting an early demise.

Among Steinem’s detractors was President Richard Nixon, who was caught on an Oval Office tape ridiculing both her and the “Ms.” title. The ridicule was mutual: In a 1972 National Press Club speech, Steinem called Nixon “the most sexually insecure chief of state since Napoleon.”

It was just one example of Steinem’s trademark wit. Before she married at 66, she often explained her singlehood by saying “I can’t mate in captivity.” Laughter, she wrote, was “the only free emotion — the only one that can’t be compelled.”

Steinem led an unconventional childhood

Gloria Marie Steinem was born on March 25, 1934, in Toledo, Ohio. Her father was a traveling antique salesperson; she spent her early years on the road with her parents, not attending regular school. “I just read everything I could get my hands on,” she told an audience at Manhattan’s 92NY earlier this year.

In 1944 her parents divorced, and 10-year-old Gloria was left to take care of her mother, a once-promising journalist who struggled to combine her work with running a household and who eventually suffered a nervous breakdown.

As a young girl she excelled at tap dancing — it was even going to be her “ticket out of Toledo,” she said. That particular career did not materialize, though, and she attended Smith College in the 1950s — a time when, she said, women were being educated to be mothers and wives.

But marriage was not on Steinem’s mind. “I was engaged to a very nice man. It wasn’t his fault, but I just didn’t want to get married,” she told her 2026 audience. Instead, she took a modest fellowship to travel to India. After that, Steinem headed to Manhattan, where she couldn’t rent her own apartment because landlords assumed if a single woman “could earn enough, you must be a hooker.”

She wanted to write about politics. But editors gave her assignments on food, fashion, and, in what she jokingly called “the lowpoint of my life,” an article about textured stockings. Even more demoralizing was her experience with an editor who, she said, gave her a choice one day: Spend the afternoon in a hotel with him, or mail his letters on the way out.

She chose to mail the letters. “There was no word for sexual harassment then,” she has said. “It was just life.”

But one assignment in 1963 sounded compelling: posing as a Playboy Bunny for an exposé in the now-defunct Show magazine. Steinem auditioned, assuming she’d be weeded out. But she was given her bunny costume — “so tight, it would give a man a cleavage” — and worked at the job for about a month.

“I could not have made a bigger mistake,” she told the AP in 2011. “It was personally and professionally a disaster. In the short term, it was much harder to get assignments, and in the long term it’s been used to ridicule me.”

Steinem’s writing became more political in her 30s

Steinem was in her 30s when she joined the editorial board of New York magazine, and her writing became more political. In 1969, she went to cover an abortion rights hearing. She had never told anyone about her illegal abortion at 22, in London. That was the moment, she has said, that triggered her career as a women’s rights activist.

Steinem founded the National Women’s Political Caucus in 1971 along with Betty Friedan, Shirley Chisholm and Bella Abzug.

Unlike her friend Abzug, however, Steinem was not comfortable in front of a crowd. Gradually, with her early speaking partner Dorothy Pitman Hughes, a Black feminist and child welfare advocate, she found her footing. The two formed a powerful partnership at a time when feminism was viewed as largely a white, middle-class movement. In an iconic photo, the two raise their right arms together in the Black Power salute.

As she grew more famous, Steinem drew attention not just for her work but for a glamorous Manhattan life. Asked once about her streaked hair and aviator glasses, she explained that the hair was inspired by Holly Golightly, the free-spirited character in “Breakfast at Tiffany’s.”

The glasses? They were “more about hiding,” she said.

When reaction was hostile, Steinem would try to reason that it was a step above ridicule. But sometimes it stung, like a snarky Esquire profile in the early ’70s, so hurtful that she cried, and colleagues called a news conference to defend her.

Steinem co-founded Ms. Magazine and stayed with it as an adviser

Then came Ms. Magazine in 1972. “I was so obsessed with the fear that it would fail and be a disgrace to the movement,” Steinem said later. And opponents hoped that would happen. TV newsman Harry Reasoner, an especially tart critic, later apologized on air. Steinem continued with Ms. until it was bought in 2001 by the Feminist Majority Foundation, and remained an adviser.

Steinem never had children, and said she never regretted it. “No, not for a millisecond,” she told the AP. “It’s important that some of us not have children, to show that we have a choice.”

In 2000, Steinem made a choice she’d never planned to: She got married, at 66, to British activist and businessman David Bale, whom she met at a fundraiser. Bale was the father of actor Christian Bale.

“I thought, ‘I haven’t changed, marriage had changed,’” Steinem said. “David and I wanted to be together, we loved each other. He needed a green card.” They married at a Cherokee ceremony in Oklahoma. Several years later, he became ill with brain lymphoma, and Steinem cared for him until his death in 2003.

Steinem said she had few regrets in life, but did miss being present when her father died. She received word that he’d been in a serious car accident, but did not respond in time, and he died alone. “I had taken care of my mom as a child, and I feared I’d never come back,” she explained. She was also sorry that she didn’t become closer to her mother. “I was so fearful of becoming her,” she said in an HBO documentary, “Gloria: In Her Own Words.”

Steinem’s many books included ‘A Life on the Road’ — though she never learned to drive

Steinem’s books included the 2015 road memoir “A Life on the Road” — but she wryly noted at the time that she’d never learned to drive. It would have taken away, she said, from her ability to converse with hosts who picked her up from airports or train stations. And, she noted, she always loved to chat with taxi drivers.

On Jan. 21, 2017, when hundreds of thousands of women gathered in Washington in their pink “pussy hats” for the historic Women’s March, it surprised nobody that the keynote speaker, and the most rousing, was Steinem.

“We are linked,” she told the crowd. “And this is a day that will change us forever because we are together.”

For many women, Steinem was the galvanizing voice that day. She would go on to advocate fiercely. But, though many saw her as a singular figure in history, she often liked to say that if she hadn’t come along, someone else would have accomplished the same things.

By the same token, she said, it was less important that younger people listen to her, and more important that they listen to themselves.

“The primary thing is not that they know who I am,” she said, “but that they know who THEY are.”

This story was originally featured on Fortune.com

The new process, though, also means that part of the AI model’s “chain of thought,” or the reasoning steps it is taking, are not expressed in natural language, making it much more difficult for humans to monitor what the model is doing and why.

Chain-of-thought monitoring is currently one of the methods companies use to make sure AI agents are not taking unintended or unauthorized actions.

Tech publication The Information first reported on OpenAI’s use of recurrent depth in Astra earlier this week. Jakub Pachocki, OpenAI’s chief scientist, and several other OpenAI researchers criticized the publication for sparking undue alarm among AI safety researchers, saying that it had limited the extent to which the looped Transformer architecture is used so that the model’s reasoning remains legible.

In response to The Information’s report, Pachocki wrote on social media platform X that “we care deeply” about chain-of-thought monitoring and that “OpenAI has worked to preserve and utilize chain-of-thought monitoring since our very first reasoning models.” Pachocki said OpenAI would share more details of Astra’s architecture in the future.

Pachocki said that he thought chain-of-thought monitoring could grow more challenging, but that this would be “for reasons not contingent on architecture changes” such as what OpenAI has begun implementing with recurrent depth. “But there are things we can do to strengthen it, and it’s a core goal of our current research program,” he wrote.

Prior to Pachocki’s statement, Steven Adler, a former OpenAI safety researcher who now runs Guidelight AI Standards, a nonprofit organization that works on AI safety benchmarks and standards, wrote on X that, if the Information report was true, “OpenAI seems to be violating one of the few redlines that exists in the AI industry.”

Peter Wildeford, policy director at the AI Policy Network, a Washington, D.C., think tank dedicated to advancing policies to help Americans prepare for the impact of AI, told Fortune that OpenAI’s use of recurrent depth was “potentially very concerning” and “potentially reckless.”

He noted that one of the only ways that OpenAI and outside AI evaluation companies were able to piece together what happened during the July incident in which several of OpenAI’s AI models autonomously attacked the company Hugging Face was by reading the models’ chains of thought. OpenAI has also said it is investing significant resources in real-time chain-of-thought monitoring going forward as part of its effort to prevent future rogue AI incidents. “If OpenAI is indeed shifting away from this, this is the wrong direction,” he said.

Several AI safety experts said they were alarmed by OpenAI’s use of looped Transformers in Astra not because they necessarily made Astra’s own reasoning more difficult to monitor. Rather, they feared that OpenAI’s move would normalize the technique, setting a precedent that other AI companies were likely to follow and expand on, eventually resulting in AI models whose reasoning steps were completely opaque to humans.

Daniel Kokotajlo, a former OpenAI governance researcher who now runs the AI Futures Project, a nonprofit research organization dedicated to forecasting the possible impacts of advanced AI, responded to Pachocki that “even if OpenAI doesn’t go further [in using architectures that make a model’s chain-of-thought less legible], others might.”

Kokotajlo urged Pachocki to lead efforts to create an industrywide standard on chain-of-thought monitorability “either to arrest the slide into oblivion or better yet to race to the top. I think this is something where we need more than just political will, we need thoughtful technical specifications.”

In a normal Transformer, tokens—the units of information that large language models process—are passed from lower layers of a model’s neural network to higher layers sequentially, with each layer applying mathematical operations in turn. In a language model, the final output of this process is text that people can read. In a reasoning model, the model is told to output a series of steps for arriving at an answer, and this text is first written to a scratch pad, which forms the model’s “chain of thought.” This intermediary text is then fed back through all the layers again to create the next steps, and so on, until the model reaches its final answer and outputs the answer text.

A cheaper but more inscrutable process

In a looped Transformer, tokens are fed multiple times through a single block (which can actually consist of several different layers, but not the entire network), with the same mathematical operations being applied each time. Critically, the output of the block is fed back into the block without its output being written to a scratch pad each time. So the method does not create a natural language “chain of thought,” even though each pass through the block does represent the model’s reasoning. AI researchers sometimes refer to the block’s output as “neuralese” since it can be processed and reasoned over by the AI model, but is not intelligible to a human. The only natural language output of the model is the final answer.

The advantage of looped Transformers is that by using the same mathematical operations each time and not having to pass every token through every layer of the network, they can wring greater performance out of a model of a given size and use less computer power. Studies have shown looped Transfomers can achieve the same performance as standard ones while using 50% to 90% less computing power. That matters for cost—a key consideration for many enterprise AI customers, who have been complaining about skyrocketing AI bills.

Obscuring part of a model’s chain-of-thought reasoning can also make the model harder to distill—a method where a smaller model is trained on the outputs of a larger model. The U.S. government and American AI companies have accused Chinese AI companies of engaging in campaigns to distill frontier AI models, including copying the chain-of-thought reasoning steps.

Looped Transformers are just one of several new architectures researchers are considering to make AI models more efficient but which can destroy all or some of a model’s chain of thought. Another technique is Prefix Sliding, which I wrote about in Tuesday’s Eye on AI newsletter. That method involves preserving the initial prompt and the last several thousand reasoning tokens, but discarding many of the intermediate reasoning tokens.

Fortune senior reporter Beatrice Nolan contributed to this story.

This story was originally featured on Fortune.com

The report from the Government Accountability Office found delays and holes in how the Secret Service updates its policies regarding threats, and arrives after a few turbulent years for the agency charged with the president’s protection.

The would-be assassin in Butler positioned himself on a roof left unsecured, nicking the president’s ear with a bullet. Months later, a man with a rifle got surprisingly close to Trump at his West Palm Beach golf course. And in April, an armed man got beyond security barriers at the White House Correspondent’s Dinner, where the president sat.

The report found that between 2015 and 2025 the Secret Service dealt with 83 security incidents and that they updated their protection policies in response to 25 of them. Among the incidents was a drone that made contact with President Barack Obama’s motorcade in 2015 and another flown about 200 feet (60 meters) over a rally for then-presidential candidate Bernie Sanders.

Failing to document why the Secret Service decided not to change their policy is the concern, said Nathan Tranquilli, acting director of the Government Accountability Office, adding that the drone incidents were a “compelling example” of that.

“Some of the missing information has been relevant to subsequent attacks,” the report read. It cited the Butler incident, where the culprit flew a drone for 11 minutes over the crowd, which helped him position himself to get clear shot at Trump.

The Secret Service also failed to update eight of 22 protection policies within a required time frame of four years. A memorandum of understanding between the Secret Service and the Diplomatic Security Service, which designates each agencies’ responsibilities for the president’s overseas security, hasn’t been updated since 1991, even though an annual review and update is required. As a consequence, the memorandum doesn’t address newer threats, such as drones.

“When you look at the Secret Service and you look at their mission, really it’s a zero fail mission, and they’ve got a ton of challenges,” said Tranquilli. “When decisions were being made about where to put time and energy, some of these things fell to the side, and, as a result, there were some delays.”

The report recommended three fixes, including that the Secret Service revise its policy to require that, when a security incident doesn’t warrant a policy update, that the rationale is documented.

A spokesperson for the Secret Service did not immediately respond to a request for comment, but the report stated that the Department of Homeland Security, which oversees the Secret Service, agreed with all recommendations and plans to implement changes.

This story was originally featured on Fortune.com

• Nvidia will purchase open-source AI platform Hugging Face for $13B
• OpenAI debuts GPT-6 Astra model
• Google debuts Gemini 3.8 Flash model
• ChatGPT experiences second outage in three days—and Claude, Grok also go down

Reporting on AI every single day has admittedly made me bit numb to the astronomical sums of money flowing into the industry, but today my eyes did a cartoonish pop when I saw how much Nvidia is buying Hugging Face for.

“Nvidia has agreed to acquire Hugging Face for $12,930,300,000,” a blog post said. While it’s not the biggest sum we’ve seen in the AI industry, it’s quite the startup exit for Hugging Face. The company was founded in 2016 by three French entrepreneurs living in New York City, who named it after the 🤗 emoji. Over the next 10 years, it has grown into a trusted repository for AI developers.

“It’s been a wild ride for Hugging Face,” cofounder Thomas Wolf said on X. “We certainly did not anticipate, back in 2016, as a tiny team of scrappy underdogs, that the field would grow so much or that the impact we could have on it would become so massive.”

The deal also makes clear how important open-weight models, like the ones Hugging Face hosts on its platform, are for the future of Nvidia’s business. As OpenAI, Microsoft, Amazon, Meta, and other closed-source AI heavyweights that have been major customers of Nvidia develop their own chips to reduce reliance on Nvidia GPUs, the company sees open-weight models as a core future revenue stream.

For Hugging Face, getting bankrolled by Nvidia could mean major growth for the platform. It currently has 18 million individual users and 200,000 enterprise clients, Nvidia told reporters today. “Our goal is is to get to 100 million AI builders in the next few next few years, and we think with the support of of Nvidia, we have more chances to get there faster,” Hugging Face CEO and cofounder Clément Delangue said.

But while Nvidia and Hugging Face line their pockets, the individuals that created and listed models on Hugging Face haven’t seen a dollar.

“It’s unfortunate [Hugging Face] never found a way for us to monetize our creations,” Eric Hartford, creator of an open-weight model called Dolphin, Chief Scientist of lazarusaie.com, and the founder of quixi.ai tells Fortune. “If I got a dollar for every download of Dolphin I’d be rich.”

Hugging Face turned down Nvidia in the past. Why now?

Nvidia has had its eye on Hugging Face for years. It participated in the company’s 2023 Series D funding round, which valued Hugging Face at $4.5 billion, TechCrunch reports. Last year, Hugging Face turned down a $500 million investment offer from Nvidia that would have valued it at $7 billion, because the investment would have made Nvidia the company’s largest minority shareholder and the founders wanted the platform to keep its independence, according to The Financial Times.

So, why did they agree to the deal now? In a briefing with reporters today, Delangue did not confirm the $500 million offer, and said previous reporting on the company’s fundraising can be “quite far from reality” since Hugging Face never publicly comments on it.

“The truth is that throughout the life of Hugging Face, we always got quite a lot of offers for acquisitions that we turned down in the past,” Delangue said. “I think this summer the planets aligned…we increasingly were convinced that Nvidia would be the perfect home for us, and so our confidence grew tremendously, and we decided that it was the best thing to do for Hugging Face, but also for the field and for AI in general.”

He’s likely referring in part to a letter Nvidia CEO Jensen Huang signed on the importance of open-weight models, which became his first X post. “Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty,” the letter said.

Also this summer, Hugging Face gained more mainstream name recognition after OpenAI’s models hacked into its repositories during training. I have to imagine that boosted the company’s brand recognition, creating the perfect storm to coax a big offer out of Nvidia.

Can Hugging Face stay neutral, or will its models now favor Nvidia hardware?

Now, the big debate in the AI industry is whether Nvidia’s acquisition will subtly influence the open-source ecosystem to favor its hardware. Model creators could choose to build their models for Nvidia hardware, especially if that makes it easiest to launch them on Hugging Face and for others to use them.

Nvidia said Hugging Face “will remain an open platform” in which developers can freely choose the models, frameworks, and computing platforms that they want. “Nvidia compute will not be required to build on or deploy through Hugging Face,” the company said.

However, some say there’s already a suspicion that Hugging Face is subtly biased toward Nvidia, and the acquisition is likely to accelerate that tilt.

“Hugging Face runs a giant library called Transformers, which is the gatekeeper for model releases,” Hartford said. “Though they have had a natural bias for Nvidia due to its overwhelming popularity, they haven’t been unnaturally biased, or mis-incentivized. With Nvidia ownership that all changes. They will still be the gatekeepers, and they will be unnaturally incentivized to favor Nvidia and disfavor Nvidia’s competitors. Not because of a conspiracy theory, but [because of] the dynamics of ownership.”

Nvidia also offers its own open models and data on Hugging Face, and says it’s “the largest contributor” on the platform, offering “more than 500 models” and “more than 250 open datasets.” It could advertise these models on Hugging Face now, as the platform owner, and gain more users.

On a more optimistic note, all of this could be a good thing for the business world. Open-weight models offer more customization and transparency than closed alternatives from the likes of OpenAI or Anthropic. It’s possible this acquisition makes it easier for organizations that are open-source-curious to finally take the plunge, opening up new ways to use AI.

With that, here’s more AI news.

Emily Forlini
emily.forlini@fortune.com
@emilyforlini

This story was originally featured on Fortune.com

Lavrov’s announcement came after Berlin blamed Moscow for an attempted drone attack at Leipzig/Halle Airport on Aug. 4. The explosives-laden drone was found near a Ukrainian plane at the airport, a major international freight hub used for providing support to Ukraine. The drone was later defused.

Earlier Thursday in Munich, several incendiary devices were thrown at a construction site near a building where several companies, including from the defense sector, are located, the Bavarian State Criminal Police Office said. There was a brief fire that caused no significant damage, police said, adding they suspect the target was one of the companies in the area.

Two Bulgarian nationals were arrested, police said, but their suspected involvement was not confirmed.

Western officials have repeatedly accused Moscow of conducting a campaign of sabotage and disruption in their countries with the aim of undermining support for Ukraine and destabilizing European nations, something Russia denies.

Germany says it has become a “daily target” of the Kremlin’s hybrid campaign.

It said Wednesday it was closing both the Russian Consulate in Bonn and the Russian House cultural center in the German capital.

Lavrov countered that cultural centers belonging to Germany’s Goethe-Institut would be closed in Moscow, St. Petersburg and Yekaterinburg, adding that any other response would show a lack of “self-respect” on Russia’s part.

He told an economic forum in the Far East city of Vladivostok that Germany “could not have failed to understand” that closing the center in Berlin would “signify the end of their cultural presence in Russia.”

Russian President Vladimir Putin said Berlin wants to distract from its own failings and dissatisfaction among voters. Any proof being offered “has been planted,” Putin said.

Europe increasingly views itself as facing what it calls hybrid attacks by Russia, marked by cyberwarfare and disinformation as well as a conventional military threat. Each side accuses the other of state terrorism.

The tension has continued for months and grown into a major foreign and defense policy challenge for countries in the European Union.

Putin is testing the West with tactics that fall “just below the threshold of war” and is determined to “harass NATO members,” the head of Britain’s MI6 spy agency said last year.

European countries seek to counter Russia’s threat

Following Germany’s claims, Poland, Denmark, Finland, the U.K., the Czech Republic and Lithuania have summoned Russian diplomats stationed in their countries.

In Norway, authorities on Wednesday seized a Russian vessel in the Arctic archipelago of Svalbard at the request of Ukraine’s state-owned Naftogaz energy group.

A Norwegian court ordered the seizure as part of efforts by Naftogaz to recover a $4.22 billion arbitration award stemming from Russia’s 2014 illegal annexation of Crimea and its seizure of Ukrainian energy assets, including gas fields and pipelines. After Moscow refused to pay the 2023 award, Naftogaz petitioned national courts for orders to seize Russian commercial assets.

Putin on Thursday compared it to “state terrorism,” and Russia’s Foreign Ministry summoned Norway’s ambassador to Moscow, Heidi Olufsen, to protest the seizure of the vessel, which it said belongs to the state weather and environmental monitoring agency.

It carried workers, scientists and cargo to Russian settlements on one of the islands in Svalbard, the ministry said, adding that Moscow “will seek to lift the vessel’s arrest.”

In Finland, meanwhile, more than 2,000 people took part Thursday in the biggest civil defense exercise in Europe since the end of World War II. Among the threats envisioned were missile strikes from neighboring Russia.

EU weighs how to respond further

EU ministers on Wednesday weighed how best to respond to the attempted drone attack but struggled to find new ways to meet the challenge without resorting to military means.

The EU has slapped almost two dozen packages of sanctions on Russia over the war, targeting over 3,000 officials, oligarchs and entities including banks, energy companies and drone makers.

One senior EU official said the bloc’s anti-terrorism procedures could be put to effective use.

“If we list individuals as terrorists, we can deploy a range of instruments to actually target those individuals and to track them, all their activities and people associated with those individuals,” EU Counter-Terrorism Coordinator Bartjan Wegter said.

He told The Associated Press that EU methods for combating terrorist financing and money laundering, as well as boosting the resilience of infrastructure like airports, could be put to good use.

“It is a very worrying development,” Wegter said of the failed Leipzig drone attack. But he noted that “it is not new what we’re seeing.”

Both Russia and Iran have used hybrid warfare methods before, he said.

“They’re actually deploying terrorist tactics” that are meant to “destabilize our society,” Wegter added.

Denmark sounds alarm about Russian sabotage plans

Danish security services are warning that Russian acts of sabotage against the country’s defense industry may be imminent.

Intelligence services have seen signs of specific planning by Russia to conduct sabotage operations in Denmark, Emil Græsholm, head of counterintelligence at the national security and intelligence service PET, told the Berlingske newspaper.

The targets are primarily companies in the defense industry and especially those associated with delivering military components to Ukraine, he said.

Russian Ambassador Vladimir Barbin said PET has not presented any concrete evidence that Moscow was preparing sabotage acts in Denmark. He said Græsholm’s comments were actually an acknowledgement of Denmark’s involvement in Ukraine.

The transformation of Danish territory into a production and logistics base for the Ukrainian military posed a security risk to Denmark, he added.

Italy’s Defense Minister Guido Crosetto told AP on Wednesday that the German airport incident was “a very serious act that could have very grave consequences.”

But he said the threats aren’t discouraging Italy from further helping Kyiv as Rome weighs whether to send SAMP/T air defense batteries. As in other Western countries, limited weapons production capacity is hampering efforts to supply more to Ukraine, he said.

___

Davies reported from Manchester, England.

___

Follow the AP’s coverage of the war in Ukraine at https://apnews.com/hub/russia-ukraine

This story was originally featured on Fortune.com

The Treasury Department proposed the change Thursday in a new regulation that, if made final, would kick in after May 2027. The rule is broadly aimed at ending any policies or programs that help students because of their race, and it specifically says such benefits in admissions, scholarships and facilities “would be incompatible” with the rule.

It’s the latest attempt by the Trump administration to pressure schools and colleges to weed out diversity, equity and inclusion policies that had become common before President Donald Trump returned to the White House with a promise to end them. Trump officials have used Civil Rights-era laws to unwind the policies, saying they discriminate against white and Asian American students.

Higher education leaders criticized the proposal.

“The administration’s latest rules changes are its most blatant attack to keep working class Americans and people of color from accessing higher education and a better life,” Mike Gavin, Alliance for Higher Education president and CEO, said in a statement. “By claiming that efforts to increase fair opportunity for all students are discriminatory, the administration is trying to gaslight the American people into believing that up is down and black is white.”

Thousands of colleges could be affected

Scores of universities have shut down or rebranded their DEI offices and ended scholarships and clubs designed for minority students under pressure from the White House. In a statement announcing the proposal, Treasury Secretary Scott Bessent suggested that even policies that are no longer under the banner of DEI could be targeted.

“Schools rebranding race-based preferences as equitable, inclusive, or diversity-enhancing does not change their discriminatory nature,” Bessent said.

The Treasury Department and IRS estimate that up to 18,000 private schools, colleges and other education institutions could be affected by the proposal.

America’s private universities have been exempt from many taxes for more than a century because they provide a public good. The benefit saves many universities millions of dollars every year.

Trump has seen the tax-exempt status as a lever to pull in his pressure campaign against colleges that he describes as hotbeds of “wokeness.” He threatened to cut the benefit for Harvard University last year during his battle with the nation’s oldest college. In a response, Harvard officials said there was no legal basis for doing so and argued it would force cuts to financial aid and crucial medical research.

In the 1970s, a college lost tax-exempt status over a ban on interracial dating

It’s incredibly rare for the federal government to go after a college’s tax-exempt status, but there’s one notable precedent. Bob Jones University, a small Christian school in South Carolina, lost the benefit in the 1970s over a ban on interracial dating and marriage on campus. The Supreme Court upheld the IRS’s decision to deny the school its exemption. The school has since ended the ban and regained tax-exempt status in 2017.

Laws forbid the IRS from targeting individuals and organizations for ideological reasons, and federal officials are not allowed to direct IRS investigations.

To maintain nonprofit status, which allows donations to be tax-deductible, organizations must follow IRS rules on lobbying, political campaign activity and annual reporting requirements, as well as other obligations.

The Trump administration describes the new proposal as a move toward restoring merit in the nation’s education systems.

Marjorie Hass, president of the Council of Independent Colleges, said the change would most likely impact donations, which are often earmarked for scholarships.

The uncertainty around the proposal also will create challenges for schools, said Tim Powers, a vice president of the National Association of Independent Colleges and Universities.

“Our institutions are committed to complying with applicable civil rights laws and maintaining safe and supportive campuses free from discrimination,” Powers said. “However, the proposed rule may create new compliance burdens and legal uncertainties for institutions of higher education that are already operating in line with existing nondiscrimination rules.”

The Justice Department has separately opened investigations into several medical schools that it accuses of favoring Black and Hispanic students in admissions. Trump officials say any such favoritism violates Title IV of the Civil Rights Act of 1964, a federal law that forbids discrimination in education and was created to fight segregation and its impact.

A statement from IRS Chief Executive Officer Frank J. Bisignano said private schools that promote discriminatory practices will no longer be exempt from taxes.

“Today’s proposed regulations put institutions on notice and schools that continue to engage in racial discrimination should expect to lose that status,” he said.

___

Associated Press Education Writers Annie Ma and Heather Hollingsworth contributed to this report.

___

The Associated Press’ education coverage receives financial support from multiple private foundations. AP is solely responsible for all content. Find AP’s standards for working with philanthropies, a list of supporters and funded coverage areas at AP.org.

This story was originally featured on Fortune.com

The startup, best known for helping users create social media graphics, is trying to carve out space in a market long dominated by Microsoft Office and Google Workspace. 

“Productivity and creativity have been separated for a long time, for no really good reason,” says Cameron Adams, Canva’s cofounder and chief product officer. Incumbent products are “a bit old-school,” he adds, arguing that they don’t match how people now produce strategy documents and pitch decks.

Still, Canva’s push into productivity software comes amid a delicate transition to offer more AI services. These new services, more popular than expected among Canva’s users, have driven up costs and forced the startup to cut its revenue growth forecast to 20%.

Canva debuted Visual Suite, its suite of productivity tools, in 2022, and overhauled it in April 2025 to add spreadsheets and coding tools. Docs is now the company’s third-most-popular design type, with adoption more than tripling in the past three years.

Mobile is becoming an increasingly important platform for Canva. More than half of Canva presentations created in Southeast Asia start on a phone, rising to three in five in the Philippines and nearly two in three in Indonesia. 

Seventy-five percent of Indonesian users access Canva through a mobile device, according to Adams, who notes that mobile-first markets in Asia and Africa see the phone as a productivity device far more than users in the U.S. or Europe might. “There are jobs that people skew more towards on mobile than they do on desktop,” he says. “We see a lot of people commenting on documents on mobile, so we privilege that sort of behavior on the mobile device.”

AI is making the phone a more credible work machine, allowing users to suggest changes verbally without needing to resort to a phone’s tiny keyboard. “Voice is a lot more usable in an AI format: I can quickly hit the microphone and spit out my thoughts. I don’t really have to structure it much,” Adams says.

That matters all the more in markets where many users are on weaker Android phones. “Performance becomes critical, particularly in Asia,” Adams says. “We’ve had a whole heap of engineers on that to make sure that Canva is a quick and stable experience when people use it on their phone.”

Canva has broader AI ambitions. The startup now has the world’s third-most-used AI web product and the fourth-most-used AI mobile app, according to calculations by venture firm a16z in March. It released Canva AI 2.0, a suite of agentic tools that build designs from conversational prompts and connect to apps like Gmail and Slack, in April.

Still, Canva has struggled to keep the AI costs to a manageable level. CEO Melanie Perkins admitted to Fortune last month that Canva “needed to reduce the cost of completing an AI task” before rolling out features more broadly to its users. 

Canva’s last known valuation was $42 billion, following an employee share sale in August 2025. Chief operating officer Cliff Obrecht told Fortune in April that shares were still trading around that level. He separately told Australian outlet Capital Brief that Canva was targeting a 2027 IPO as it built out its AI services.

For Adams, however, Canva’s aim is not to use AI to eliminate the work of creating and refining a design. “If you can ask AI to create an image, a video, or a new design for you that’s higher quality than what you could achieve before, that’s a win,” Adams says. But he argues users can’t simply “hit send” on what AI produces: They need a platform to “dive in, move things around, change the words” before passing it to teammates.

The main point, Adams says, is to bring the “human element into it,” and move Canva “beyond AI slop.”

This story was originally featured on Fortune.com