Inside the Claude Mythos 5 battle

As the rest of the world celebrated the United States’ first World Cup win against the New York Knicks, Anthropic spent the week taking on the Trump administration in its latest release. At 5:21 PM on Friday, the company received US instructions to suspend access to its Mythos 5 and Fable 5 AI models by “any foreign entity” inside or outside the US, “including Anthropic’s foreign employees.” The only way that was possible, Anthropic decided, was enough stop it medicine was destroyed last week hyping — is heading to Washington, DC in hopes of changing President Donald Trump’s mind. Now, in the coming days, the US government may make major changes to the way all companies operate, which will have a major impact on the American AI industry.

Claude Mythos 5 and Fable 5 were built on the same foundation as Anthropic’s Mythos Preview, which Anthropic called the most dangerous to release openly. (The company’s warning could be seen as a real concern or a serious reaction to their model – or both.) The Mythos 5 was given to a select group of government agencies and companies, while the Fable 5, which had additional protection, was considered “safe for use.” But when a report indicated that the defenses had failed, Anthropic’s dire warnings about the Mythos falling into the wrong hands returned.

A source familiar with the situation, who was involved in the negotiations between Anthropic and the Trump administration, said that the administration called the AI ​​lab on Friday around 1 pm ET and gave the company 90 minutes to close access to Mythos 5 and Fable 5. If it did not, then the government will impose commercial regulations on Anthropic and the authority of the US Department of Anthropic.

The source said that Anthropic officials spoke with the White House within 15 minutes of the first call, confirming that CEO Dario Amodei joined the conversation about an hour and 15 minutes after the first call. Amodei spoke directly with US Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber ​​Director Sean Cairncross, on several occasions, the source confirmed.

Anthropic wrote in a release Friday that the company believes the government “believes it has identified a way to bypass, or ‘jailbreak’ Fable 5.” Rather than a threat, however, Anthropic said the prison in question was a “narrow, inaccessible facility” that was “shared with the government” by an organization the company declined to name. Also, Anthropic said that this behavior was not unique to Fable 5. “We have reviewed the report that we believe is the basis of the government’s guidance and confirmed that the level of skill shown there is widely available from other models (including OpenAI’s GPT-5.5),” Anthropic wrote.

Street lights reportciting another well-known source, that the unrest started because the US government was concerned that a group allied with China had acquired the technology. But the source said that the Chinese rumors went back a few weeks, about a major international communications company that was approved to be included in the Mythos Preview, and that after the US government shared its concerns, Anthropic blocked the opportunity.

An X post and David Sacks, the former US government AI and crypto czar who stepped down in March, did not mention China. Sacks, however, referred to an unnamed group that leaked the story to the government, calling it “a trusted friend of Anthropic and the USG who was testing Fable (which) came up with the jailbreak of those guards.”

Others reports The head of Amazon Andy Jassy is the person who expressed concern to the US government after the researchers at Amazon made a red team Fable 5. This is in conflict with other independent players, who took action. he said he was impressed by the level of security.

A source familiar with the negotiations said Amazon’s research was clearly mentioned in discussions with the US government. The person added that Anthropic had access to the paper within days of Friday’s export control and has been going back and forth since then with Amazon researchers to discuss it.

Everything on the site, the source said, can be implemented with OpenAI’s GPT-5.5.

Anthropic spent the weekend scrambling to make good with the Trump administration, starting with virtual meetings and then flying staff to DC, including Dave Orr, Anthropic’s chief security officer; Logan Graham, who runs his red team and has led work on Project Glasswing; and Nicholas Carlini, a leading frontiersman and cybersecurity researcher. Axios reportciting a source familiar with the Trump administration’s thinking, that the company has made repeated mistakes in communicating with management and that it “has not been successful in trying to talk to management and appreciate differences of opinion.” For Anthropic, the timing could not be worse: the company invested in Mythos help to recoverin part, because of months of public debate with the United States Department of Defense.

A source familiar with the negotiations said that Anthropic had previously explained the authority on Fable 5, and that the US Department of Commerce had tried to deploy it, with no concerns shared at the time. The source added that Anthropic has been working closely with government agencies since the release of the Mythos Preview.

The Trump administration initially took a defensive approach to AI — but after Mythos, it has to be misunderstoodeven lamenting the fear of losing the AI ​​race to China. Now, prominent cybersecurity leaders have said so he warned that leaving Mythos 5 and Fable 5 could give China a big AI advantage. Trump’s move is done international invitations instead of American AI systems, effectively putting a new type of US AI industry on ice.

A public letter from the technical and cybersecurity authorities requested that restrictions on Fable 5 be lifted on Sunday. “Not all of us agree that AI legislation is the right way forward,” the letter says, adding that if legislation is passed regardless, it should be based on “scientific analysis made up of ideas from industry and academia.”

Alex Stamos, head of sales at Corridor, said Seaside prepared a public letter because the number of threats in the last decade, including, written in many different languages, requires AI to identify the bad guys before they find them. “We’re in a race, and I think policymakers don’t understand,” Stamos said. “There’s an incredible arrogance, the idea that American labs are so far ahead of our enemies that it’s always going to be true, that it’s necessary to deny access because of it. I just think it’s ridiculous. If the labs are ahead, it’s months. you know.”

The public letter goes on to say that while Anthropic’s Mythos members are skilled at finding cybersecurity threats and exploiting exploits, they are not “exceptionally good” at these tasks and that Fable 5’s security was “so brutal that it was a source of public humor on launch day.” Stamos said Seaside that “there is a real increase in Mythos talent. Anthropic is the one who started this, clearly… Mythos is good, but the real change was last year.”

Stamos said the industry is overwhelmed with backup agreements being signed with non-US companies and heavy-duty models being used on other hardware systems because last week made political threats more part of the company’s plans than ever before.

“They’re laughing at us in Beijing right now,” Stamos said. “One of America’s geniuses is being swayed by the United States government while we’re competing with the Chinese people. It’s ridiculous. That’s why I wrote the letter, and I think that’s why a lot of people signed the letter.”

Ben Van Roo, co-founder and CEO of Legion Intelligence, a national security advisory group, said Seaside that “the instruction that ‘no foreigner should use this model’ is something that cannot be followed.” He added, “When I read this, the whole … (network of) AI community is exploding.”

To make things even faster, OpenAI, Google, and Microsoft have all come out with their own products similar to Anthropic’s Mythos, making many similar claims about their performance and risk. If the Trump administration bans advanced models of Anthropic cybersecurity, it could also ban competitors’ models. This could encourage AI industry leaders to join forces with Anthropic or, as in the battle for autonomous weapons with the Pentagon, position themselves as a group. safe and legal method.

Even if the Trump administration is trying free technology companies As for the legal issues, the Anthropic system could be a huge deterrent to powerful AI models – depending on how the next few days play out.

Legion Intelligence’s Van Roo called it an “unselected phase” in the regulatory process, adding that he doesn’t think this is the last time it will happen.

We’ve also entered an era of AI populism, where more and more people are pushing back against the power of AI companies and the increased power overhead through data center demonstrations, promises to give up the use of AI chatbots, wrongful death lawsuits, and attempted attacks on AI company CEOs. Van Roo says the Trump administration’s recent actions against Anthropic could create “great fear and anxiety, perhaps for the wrong reasons.”

A source familiar with the negotiations described the weekend’s talks as constructive, with some members of the administration agreeing that imposing export controls on model suppliers is not appropriate, as competitors with similar products could be subject to the same restrictions – and since the US government is doing well. they are currently investigating program that would encourage the export of American AI systems.

As Anthropic continues to negotiate with the US government, there is a small chance that many other issues between the company and the Pentagon will not come up – for example, the ongoing battle between Anthropic and the Department of Defense over the legal framework for the use of Anthropic technology by the US military.

“This is new and we’ve never had anything that could be this complicated, and it has real consequences” in terms of how we can take advantage of powerful samples, Van Roo said. “Who can use this new technology that is beyond our control?”