U.S. export – A U.S. Commerce Department enforcement letter sent to Anthropic on Friday invoked an export control directive that barred non-Americans—including Anthropic employees—from accessing its Fable 5 and Mythos 5 models. The letter was not made public, and the compan
On Friday afternoon, the U.S. Commerce Department sent Anthropic an enforcement letter that didn’t just raise compliance questions—it changed what customers could use before the weekend ended.
The letter invoked an obscure export control directive banning non-Americans. including Anthropic’s own employees. from accessing two of its latest AI models: Fable 5 and Mythos 5. Anthropic said it believes the letter was related to a bypass of the model’s guardrails. but it also said it isn’t sure. because the letter doesn’t include specific details. The letter hasn’t been made public.
To keep up with the directive, Anthropic shut down both top models to all customers. In practice, that meant the U.S. government didn’t need the slow grind of court proceedings to pull advanced models offline.
The move landed with a jolt inside an industry that likes to frame AI deployments as technical, not political. During the weekend, a tense standoff between the government and Anthropic was described in terms of “personality differences,” not a technical problem with the AI products themselves.
New details about what triggered the government’s concern have only intensified doubts.
Katie Moussouris—an established cybersecurity veteran and the founder of Luta Security—said in a blog post that Anthropic recently shared with her a private copy of a paper written by security researchers describing an alleged guardrail bypass in Fable 5. Moussouris added that Anthropic reached out to ask for her take on the paper.
Moussouris’ blog post described how the researchers triggered the guardrail bypass. but said that the bypass itself “should never have triggered an export control.” Her argument rests on the difference in framing: she described the distinction between asking a model to “review code for security issues” and asking it to “fix this code.” She said the end result is largely the same. even if the prompts are posed slightly differently.
“The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense,” Moussouris wrote. She criticized the export control directive as hasty, heavy-handed, and misguided.
In the days since, Moussouris and dozens of other top security researchers and experts have called on the Trump administration to revoke the export control order. Their concern is direct: pulling advanced cybersecurity capabilities from U.S. network defenders, they argue, is “dangerous.”
The letter also echoes a long-running U.S. problem with export rules for cybersecurity tools—past administrations used language that was so broad it nearly outlawed legitimate security and vulnerability research. But in this case, the current directive appears to point in a different direction. The administration’s action has been described as retaliatory rather than purely regulatory. with critics saying the letter signals a willingness to punish firms through sudden enforcement rather than clear technical adjudication.
Justin Hendrix. editor of Tech Policy Press. said the move is “likely to raise alarms in foreign capitals about the reliability of American AI for critical applications.” His concern wasn’t about jailbreaks in isolation. It was about trust—about whether AI companies in the United States can operate without government pressure shaping what gets released and what stays locked behind rules.
The Trump administration hasn’t confirmed why it invoked the export control directive. That silence has left space for sharper speculation: whether officials misread the information and reacted too quickly. whether warnings tied to comments by Amazon CEO Andy Jassy reached senior officials in a way that triggered caution—or whether something was lost in translation. Another possibility raised in the reporting is that the administration is trying to pressure Anthropic. with whom it has a “fractious relationship.”.
To quote Hendrix. “the climate is one of a cloud of suspicion that senior officials are picking favorites based on personal and political factors.” His point lands because the consequences are immediate and visible: the government forced a tech company to pull its models offline through a swift. unilateral action that didn’t appear to require court approval.
When a decision of this scale arrives without the letter being public and without clear technical explanation, it doesn’t just freeze one release. It sets a precedent—one that critics say could become the blueprint for how Washington intends to control the release of American-made software.
This time, the pressure was applied to Anthropic. Tomorrow, it could be any other provider of advanced AI tools.
Anthropic Fable 5 Mythos 5 export control directive U.S. Commerce Department guardrails bypass Luta Security Katie Moussouris cybersecurity research AI regulation Tech Policy Press Justin Hendrix national security AI models offline
So basically the government shut off AI. Cool cool.
I don’t get it, if it’s just “guardrails” why does it involve export controls? Sounds like they’re using this as an excuse to control who can access stuff, like always.
Wait, “non-Americans” can’t access it even if they work at Anthropic?? So they’re banning employees too, right? I’m confused but also that seems like the US trying to make sure nobody outside their little circle touches it. Also “guardrail bypass” sounds like hackers, not export law.
This is wild because companies always say it’s technical, but then suddenly it’s political. They didn’t even make the letter public so how is anyone supposed to know what happened? And “personality differences”?? like the AI had attitude?? My guess is the bypass was real and they panicked, but who knows.