Last Friday, the U.S. Commerce Department invoked an obscure export control directive to force Anthropic to take its latest models, Fable 5 and Mythos 5, offline. The letter, citing unspecified national security concerns, banned non-U.S. persons—including Anthropic employees—from accessing the models. Anthropic complied immediately, shutting down both models to all customers. No court approval was required.

The trigger appears to be a security research paper, shared privately with Anthropic by Amazon researchers, describing a guardrail bypass in Fable 5. But according to cybersecurity veteran Katie Moussouris, founder of Luta Security, the described behavior—asking the model to "fix this code" versus "review code for security issues"—is a semantic distinction with identical practical outcomes. She argues the bypass "should never have triggered an export control" and that attempts to fix it would only weaken defensive capabilities.

Axios reports the action may stem from "personality differences" between Anthropic and the Trump administration rather than technical merit. Tech Policy Press editor Justin Hendrix warns the move undermines trust in American AI reliability globally, noting the government now sets a precedent for unilateral product shutdowns based on opaque reasoning.

The directive mirrors past overcorrections, like 2010s export law language that nearly outlawed legitimate vulnerability research. This time, security experts and researchers are calling for revocation, describing the pull as dangerous for network defenders. The underlying engineering lesson: guardrail bypasses are inevitable, but conflating them with export violations creates a chilling effect on model release and defensive AI research.

Source: TechCrunch