The Trump administration has forced Anthropic to disable access to its newly released Mythos 5 and Fable 5 AI models after issuing a sweeping export control directive late Friday afternoon, precipitating a crisis that could reshape the regulatory landscape for advanced artificial intelligence in the United States. The directive, received by Anthropic at 5:21 PM ET, ordered the company to suspend access to both models by any foreign national inside or outside the United States — including foreign national employees of Anthropic itself. Faced with an implementation requirement that was functionally impossible to enforce selectively, the company chose to disable both products entirely and dispatched senior executives to Washington, D.C., in an urgent bid to reverse the decision.
The 90-Minute Ultimatum and the Executive Response
The sequence of events began around 1:00 PM ET on Friday, when the administration gave Anthropic a 90-minute ultimatum: shut down access to Mythos 5 and Fable 5 voluntarily, or face formal export controls imposed by the U.S. Commerce Department. Within 15 minutes of that initial call, Anthropic executives were in contact with the White House. CEO Dario Amodei joined the discussions approximately one hour and 15 minutes later, holding direct conversations with Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross — in some cases multiple times.
By Monday evening, negotiations had concluded for the day with no resolution. The company had scrambled over the weekend, moving from virtual meetings to deploying key personnel to Washington, including Dave Orr, Anthropic’s head of safeguards; Logan Graham, who leads its frontier red team and has overseen Project Glasswing; and Nicholas Carlini, a leading frontier developer and cybersecurity researcher.
What Triggered the Government Action
Anthropic stated in a Friday release that it believed the administration had become aware of a method for bypassing, or jailbreaking, Fable 5. The company described the vulnerability as a “potential narrow, non-universal” exploit that had been shared with the government by an unnamed entity. Crucially, Anthropic argued that the reported capability was not unique to its model, asserting that the same level of functionality was widely available from other frontier systems, including OpenAI’s GPT-5.5.
This assessment was supported by an internal source familiar with the negotiations, who confirmed that every capability identified in the report that triggered the government’s action could be replicated by OpenAI’s GPT-5.5. The implication is clear: if the administration’s standard for restriction is the presence of a jailbreakable vulnerability, then every major frontier model would fall under the same criterion.
Multiple reports have pointed to Amazon CEO Andy Jassy as the individual who flagged concerns to the U.S. government, after Amazon researchers conducted red-teaming on Fable 5. The source familiar with the negotiations confirmed that Amazon’s research was explicitly referenced during discussions with the administration. Anthropic had reportedly had access to that research paper within days of the export directive and had been engaged in an ongoing technical dialogue with Amazon researchers about its findings. This narrative, however, stands in tension with assessments from independent red-teamers, who have publicly stated they were impressed with the level of protections in Fable 5.
The China Dimension and Shifting Policy Ground
Reports from Semafor, citing a source familiar with the situation, indicated that the administration’s concern originated from worries that a China-linked group had accessed the technology. Those rumors had circulated for weeks, centered on a large global telecommunications company that had been initially cleared for access to Mythos Preview. When the administration shared its concerns, Anthropic immediately revoked that access.
David Sacks, who served as the U.S. government’s AI and crypto czar until March, posted on X that the issue was brought to light by “a highly credible trusted partner of both Anthropic and the USG who was testing Fable” and who “came forward with a jailbreak of those guardrails.” Notably, Sacks did not mention China in his post.
The Trump administration initially pursued a hands-off approach to AI safety, but the post-Mythos era has seen a marked shift toward ambivalence and intervention. The administration now finds itself caught between its stated desire to reduce regulatory burdens on technology companies and a deepening anxiety about losing the AI race to China. Prominent cybersecurity leaders have already warned that sidelining Mythos 5 and Fable 5 could hand China a significant advantage in AI capabilities. Meanwhile, the move has galvanized international calls for alternatives to American AI systems, as other nations accelerate plans for sovereign AI infrastructure.
Industry Backlash and the Free Fable Letter
On Sunday, a public letter organized by Alex Stamos, chief product officer at Corridor, and signed by numerous tech and cybersecurity executives, called for the restrictions on Fable 5 to be repealed. The letter argued that if AI regulation is to proceed, it should be grounded in “scientific evaluations developed with input from industry and academia.”
Stamos framed the situation in stark competitive terms. “They are laughing at us in Beijing right now,” he said. “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.” He argued that the assumption of permanent American technological supremacy is a dangerous illusion, noting that cutting-edge U.S. models are only about six months ahead of their Chinese counterparts in open evaluations — and those are the models we know about.
The letter also contested the premise that Mythos-class models represent an unprecedented leap in capability for cybersecurity tasks. While acknowledging that these models are skilled at finding vulnerabilities and exploiting them, the signatories argued they are not “uniquely good” at these functions. Stamos noted that Fable 5’s safeguards were “so aggressive as to be the source of humor in the cyber community on launch day.”
A Precedent With No Clear Boundaries
Ben Van Roo, co-founder and CEO of Legion Intelligence, which builds agent systems for the national security community, described the directive’s scope as unenforceable in practice. “The directive of ‘no foreign national should use this model’ is the most impossible thing to enforce,” he said, noting that the AI community’s reaction was immediate and intense.
The precedent set by this action extends well beyond Anthropic. OpenAI, Google, and Microsoft have all released products comparable to Mythos, making similar claims about their effectiveness and associated risks. If the administration proceeds with banning Anthropic’s advanced cybersecurity models, it establishes a regulatory framework that could logically extend to competitors’ models as well. This could either galvanize the industry to unite in support of Anthropic or, as has happened in previous disputes over autonomous weapons and military contracts, drive competitors to position themselves as safer and more compliant alternatives.
The source familiar with the negotiations described the weekend’s conversations as constructive, with some administration officials acknowledging that imposing export controls on model providers is not an ideal approach — particularly since the Commerce Department is currently exploring a program to encourage the export of American AI systems. The tension between promoting AI exports and restricting access to powerful models is a contradiction that the administration has yet to resolve.
The Deeper Regulatory and Political Context
This confrontation arrives at a moment when Anthropic was already navigating a fraught relationship with the U.S. Department of Defense over acceptable usage policies for its technology by the military. The company had banked on Mythos to help recover from months of high-profile clashes with the Pentagon, making the timing of the export directive particularly damaging.
The administration’s action also unfolds against the backdrop of a broader cultural shift that Van Roo describes as “AI populism” — a growing backlash against the AI industry’s concentration of power, visible in data center protests, pledges to quit using AI tools, lawsuits over wrongful deaths attributed to AI systems, and even attempted attacks on AI company executives. The government’s move against Anthropic, he warned, could stoke “greater fears and concerns, potentially for the wrong reasons,” by reinforcing narratives of AI as an ungovernable threat requiring drastic intervention.
Van Roo characterized the current situation as “uncharted territory” in regulatory terms, and expressed doubt that this would be the last such incident. “Who gets to use this new technology that continues to outpace our own ability to regulate it?” he asked, capturing the central question that the coming days may begin to answer.
What This Means for the Industry Now
As negotiations continue without resolution, the immediate practical consequence is that Mythos 5 and Fable 5 remain unavailable to all users, domestic and foreign alike. The industry is already responding: backup contracts are being signed with non-U.S. providers, and open-weight models are being deployed on alternative hardware arrangements as companies incorporate political risk into their core business planning for the first time at scale.
For developers, security researchers, and enterprise customers evaluating AI procurement strategies, the lesson is that regulatory vulnerability has become a first-order consideration. The assumption that frontier American AI models will remain reliably accessible — and that the U.S. government will act primarily to support rather than constrain domestic AI leaders — can no longer be taken for granted. Any organization building critical infrastructure on top of a single frontier model provider should now evaluate diversification strategies, including open-weight alternatives and models hosted in jurisdictions outside direct U.S. export control reach.
The coming days will determine whether this action becomes an isolated intervention or the template for a new era of AI regulation. If the administration extends similar restrictions to other models and companies, the competitive landscape of global AI will shift decisively — and not necessarily in favor of the United States.
