Claude Mythos 5: The Conflict Unpacked

The US government has effectively halted the deployment of Anthropic’s latest and most advanced AI models, Mythos 5 and Fable 5. Sources close to the situation indicate that the AI research lab and its advocates spent the weekend attempting to convince authorities that Fable 5 was not excessively powerful.

While the nation celebrated significant sporting victories, Anthropic found itself in a tense standoff with the Trump administration over its new AI models. On Friday at 5:21 PM, the company received a US export control directive mandating the suspension of access to Mythos 5 and Fable 5 for "any foreign national" both within and outside the US, including its own foreign national employees. Anthropic determined that the only way to comply was to completely disable the products it had heavily promoted in the preceding week. This prompted an urgent trip to Washington, D.C., in hopes of reversing President Donald Trump’s decision. This action by the US government could profoundly reshape the trajectory of the AI industry, delivering a substantial setback to American AI companies.

Claude Mythos 5 and Fable 5 share a common architectural foundation with Anthropic’s Mythos Preview, a model the company itself previously deemed too hazardous for public release. (The motivations behind Anthropic's warnings have been debated as either genuine concern or strategic marketing, or possibly both.) Mythos 5 was initially made available to a select group of government agencies and private firms, while Fable 5, which incorporated enhanced safeguards, was considered "safe for general use." However, a subsequent report suggesting a potential failure in these guardrails brought Anthropic's prior warnings about Mythos falling into the wrong hands back into sharp focus.

A source familiar with the negotiations between Anthropic and the Trump administration revealed that the administration contacted the AI lab on Friday around 1 PM ET, issuing a 90-minute ultimatum to cease access to both Mythos 5 and Fable 5. Failure to comply, the company was warned, would result in the imposition of export controls by the US Commerce Department.

Within 15 minutes of that initial call, Anthropic executives were reportedly in discussions with the White House. The source confirmed that CEO Dario Amodei joined these high-level conversations approximately an hour and 15 minutes later, engaging directly, and in some instances repeatedly, with US Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross.

In a Friday release, Anthropic stated its belief that the government "believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5." However, Anthropic downplayed the severity, characterizing the alleged jailbreak as a "potential narrow, non-universal" issue that was "shared with the government" by an unnamed entity. Furthermore, Anthropic asserted that this behavior was not unique to Fable 5. The company wrote, "We have reviewed a report that we believe is the basis of the government’s directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5)."

Semafor, citing an informed source, reported that the controversy stemmed from US government concerns regarding a China-linked group potentially accessing the technology. However, the source indicated that rumors about China had circulated for weeks, referencing a major global telecommunications company that had initially been approved for access to Mythos Preview. When the US government expressed its concerns, Anthropic promptly revoked that access.

David Sacks, the former US government AI and crypto czar who resigned in March, also weighed in via an X post, though he did not mention China. Sacks did, however, refer to the unnamed entity that brought the issue to the government's attention as "a highly credible trusted partner of both Anthropic and the USG who was testing Fable [which] came forward with a jailbreak of those guardrails."

Some reports suggest that Amazon CEO Andy Jassy was the individual who raised concerns with the US government following red-teaming exercises on Fable 5 conducted by Amazon researchers. This perspective contrasts with certain independent red-teamers who had expressed admiration for the model's protective measures.

The source involved in the negotiations confirmed that the Amazon research was explicitly referenced during discussions with the US government. This individual added that Anthropic had gained access to the Amazon paper within days of the Friday export control directive and had been in ongoing communication with Amazon researchers to discuss its findings.

According to the source, all capabilities detailed in that particular paper could also be achieved using OpenAI’s GPT-5.5.

Anthropic dedicated the weekend to mending relations with the Trump administration, commencing with virtual meetings before flying key personnel to Washington, D.C. This team included Dave Orr, Anthropic’s head of safeguards; Logan Graham, who leads its frontier red team and Project Glasswing efforts; and Nicholas Carlini, a prominent frontier developer and cybersecurity researcher. Axios, citing a source familiar with the Trump administration’s perspective, reported that Anthropic has consistently mismanaged its communication with the administration, failing to effectively engage and acknowledge ideological differences. The timing of this crisis is particularly unfortunate for Anthropic, as the company had been relying on Mythos to help recover from months of high-profile disputes with the US Department of Defense.

The source familiar with the negotiations also stated that Anthropic had pre-briefed the administration on Fable 5, and that the US Department of Commerce had conducted pre-deployment testing without raising any concerns at the time. The source further noted Anthropic's close collaboration with government agencies since the release of Mythos Preview.

Initially, the Trump administration adopted a less interventionist stance on AI safety. However, in the wake of the Mythos developments, its position has grown more ambiguous, fueled by anxieties over the US potentially losing the AI race to China. Cybersecurity leaders have now cautioned that sidelining Mythos 5 and Fable 5 could inadvertently grant China a significant AI advantage. This move by the Trump administration has already spurred international calls for alternative AI systems to those from the US, effectively sidelining a major American AI company's flagship model.

A public letter signed by technology and cybersecurity executives on Sunday urged the repeal of restrictions on Fable 5. The letter acknowledged, "Not all of us agree that AI regulation is the right way forward," but emphasized that if regulations are inevitable, they should be founded on "scientific evaluations developed with input from industry and academia."

Alex Stamos, chief product officer at Corridor, informed The Verge that he organized the public letter out of concern that the vast number of software vulnerabilities accumulated over the past decade, across various coding languages, necessitate AI to patch them before malicious actors exploit them. "We’re in a race, and I think policymakers don’t understand that," Stamos stated. He criticized what he called "this weird arrogance, this idea that American labs are hugely ahead of our adversaries that will always be true, that it’s really important to restrict access because of that. I just think that’s foolish. If the labs are ahead, it’s only by a matter of months. And you can see that in the open evaluations. The cutting-edge models are only something like six months ahead of the Chinese models — and those are the models we know about."

The public letter further asserted that while Anthropic’s Mythos-class models excel at identifying cybersecurity vulnerabilities and exploiting them, they are not "uniquely good" at these tasks. It also highlighted that Fable 5’s safeguards "were so aggressive as to be the source of humor in the cyber community on launch day." Stamos told The Verge that there is "a real overstatement of Mythos’ capabilities. Anthropic is somewhat responsible for this themselves, clearly … Mythos is great, but the real turning point was really last year."

Stamos observed that the industry is now seeing a proliferation of backup contracts with non-US companies and deployments of open-weight models on alternative hardware. This shift, he noted, is a direct consequence of the past weekend's events, which have cemented political risk as a central component of companies' business strategies more than ever before.

"They are laughing at us in Beijing right now," Stamos declared. "One of America’s champions is being kneecapped by the US government while we’re in a race with the Chinese. It’s just incredibly stupid. That’s why I wrote the letter, and I think that’s why a lot of people signed onto it."

Ben Van Roo, co-founder and CEO of Legion Intelligence, which develops agent systems for the national security community, told The Verge that "the directive of ‘no foreign national should use this model’ is the most impossible thing to enforce." He added, "When I first read that, my whole… [network of] AI community nerds was exploding."

Adding to the urgency of the situation, OpenAI, Google, and Microsoft have all released comparable products to Anthropic’s Mythos, making similar claims about their effectiveness and associated risks. If the Trump administration imposes a ban on Anthropic’s advanced cybersecurity models, it could establish a precedent for banning competitors' models as well. This scenario might compel AI industry leaders to either unite in support of Anthropic or, echoing past disputes with the Pentagon over autonomous weapons, position themselves as safer and more compliant alternatives.

Even as the Trump administration ostensibly aims to alleviate regulatory burdens on tech companies, the order against Anthropic could paradoxically lead to significant restrictions on powerful AI models, depending on how events unfold in the coming days.

Van Roo of Legion Intelligence described the regulatory environment as "uncharted territory," expressing his belief that similar incidents are likely to occur again.

Moreover, the present era is marked by growing "AI populism," with increasing public resistance to the AI industry's expansive influence and the concentration of power at its apex. This pushback manifests through data center protests, pledges to discontinue using AI chatbots, lawsuits concerning wrongful deaths, and even attempted attacks on AI company CEOs. Van Roo suggests that the Trump administration’s recent actions against Anthropic could inadvertently fuel "greater fears and concerns, potentially for the wrong reasons."

The source familiar with the negotiations characterized the weekend’s discussions as constructive. Some administration officials reportedly acknowledged that imposing export controls on model providers is not ideal, given that competitors with similar products might face identical restrictions, and considering the US government is actively exploring a program to encourage the export of American AI systems.

As Anthropic continues its negotiations with the US government, it is highly probable that the company's existing and numerous issues with the Pentagon will also arise. These include the ongoing disagreement between Anthropic and the Department of Defense regarding acceptable usage policies for Anthropic's technology by the US military.

"This is new and we’ve never had anything potentially this drastic before, and it does have some real ramifications" concerning the enforcement of access to powerful models, Van Roo stated. He questioned, "Who gets to use this new technology that continues to outpace our own ability to regulate it?"