U.S. ban on Anthropic models wasn’t about AI jailbreaking

A recent enforcement letter from the U.S. government to Anthropic, which effectively compelled the company to take its newest AI models offline just before the weekend, should serve as a stark warning for any U.S. tech firm — whether an AI lab or not.

Here’s a quick recap: On Friday afternoon, the U.S. Commerce Department sent Anthropic a letter citing a little-known export control rule that barred non-Americans — including Anthropic’s own employees — from accessing Fable 5 and Mythos 5, citing an unspecified national security concern. Anthropic said it believes the letter relates to a bypass of the model’s safety guardrails but is uncertain because the letter lacks specific details. The letter has not been made public.

In response, Anthropic shut down both of its flagship models for all customers to ensure compliance with the directive. As a result, the U.S. government successfully forced a tech company to take its models offline through a swift, unilateral action that appeared to require no court approval.

Friday’s intervention by the Trump administration shows that the AI industry is not immune to government interference. It also sends a broader message to the tech world: comply, or we can shut down both you and your products.

Citing sources, Axios described a tense situation over the weekend between the two major parties, suggesting that “personality differences” between Anthropic and the Trump administration — rather than a technical flaw in the AI products — triggered the export directive.

New details that emerged over the weekend now further cast doubt on the government’s already questionable reasoning.

Katie Moussouris, a cybersecurity veteran, researcher, and founder of Luta Security, wrote in a blog post that Anthropic recently shared with her a private copy of a paper by security researchers describing an alleged guardrail bypass in Fable 5. (The Wall Street Journal reports that the paper’s authors are security researchers at Amazon.) Moussouris said Anthropic reached out to ask for her perspective on the paper.

In her blog post, Moussouris described how the researchers triggered the guardrail bypass but noted that the bypass itself “should never have triggered an export control.” The key distinction lies between asking an AI model to “review code for security issues” versus asking it to “fix this code.” The end result is largely the same, even when the questions are phrased slightly differently.

“The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense,” said Moussouris, who criticized the export control directive as hasty, heavy-handed, and misguided.

Moussouris, along with dozens of other top security researchers and experts, has since called on the Trump administration to revoke the export control order, calling the move to pull advanced cybersecurity capabilities from network defenders in the U.S. “dangerous.”

Previous administrations have made sweeping decisions based on knowledge gaps. For example, language used by the U.S. government during the 2010s to fix export laws covering cybersecurity tools that could also be used for cyberattacks was so broad that it inadvertently nearly outlawed legitimate security and vulnerability research.

However, the Trump administration’s directive appears retaliatory.

Justin Hendrix, editor of Tech Policy Press, said the Trump administration’s move is “likely to raise alarms in foreign capitals about the reliability of American AI for critical applications.” The message is that AI companies in the United States cannot be trusted to operate without interference from the U.S. government.

The Trump administration has not confirmed why it invoked the export control directive. Did officials misinterpret the report and overreact? Did Amazon CEO Andy Jassy say something to senior government officials that prompted the response, out of caution or spite? Was something lost in translation, or was this a way to pressure Anthropic, with whom the administration already has a fractious relationship? It is possible that the White House was unaware of the far-reaching consequences of the letter’s demand and officials are now scrambling to undo the damage they caused.

To quote Hendrix, “the climate is one of a cloud of suspicion that senior officials are picking favorites based on personal and political factors.” The aftermath is that the government has set a dangerous precedent regarding how much control it intends to exert over the release of American-made software.

This time, the government took issue with Anthropic; tomorrow, it could be with anyone else.