On Friday afternoon, the U.S. Commerce Department sent Anthropic a letter invoking an obscure export control directive that banned non-Americans, including Anthropic's employees, from accessing Fable 5 and Mythos 5. The stated reason was an unspecified national security concern. The real reason appears to be a misreading of a security paper detailing a guardrail bypass that security researchers say isn't dangerous.
Anthropic complied immediately, shutting down both models to all customers. The government didn't need a court order. It just sent a letter.
The Paper That Started the Panic
Katie Moussouris, founder of Luta Security and a veteran cybersecurity researcher, said Anthropic privately shared a paper written by Amazon security researchers describing an alleged guardrail bypass in Fable 5. Moussouris analyzed it and concluded the bypass "should never have triggered an export control." The difference? Asking the model to "review code for security issues" versus "fix this code." End result is largely the same, but the phrasing is different.
Moussouris put it bluntly: "The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense." She called the export control directive hasty, heavy-handed, and misguided. Dozens of top security researchers have since called on the Trump administration to revoke the order, warning that pulling advanced cybersecurity capabilities from network defenders is "dangerous."
A Precedent That Should Terrify Anyone Shipping Software
The move echoes the 2010s, when the U.S. government used broad language to fix export laws covering cybersecurity tools, inadvertently near-outlawing legitimate security research. But this time feels different. Justin Hendrix, editor of Tech Policy Press, said the directive "is likely to raise alarms in foreign capitals about the reliability of American AI for critical applications."
The administration hasn't confirmed why it invoked the directive. Did officials misread the report and freak out? Did Amazon CEO Andy Jassy say something to senior officials? Was it a way to pressure Anthropic, with whom the administration already has a fractious relationship? Regardless, the message is clear: the government can shut down any AI product without court approval, based on an opaque national security claim.
Tomorrow it could be anyone else's software. The only difference this time is that Anthropic tried to cooperate.
Source: The US government's Anthropic models ban was never about an AI jailbreak
Domain: techcrunch.com
Comments load interactively on the live page.