source: techcrunch ai: the us government’s anthropic models ban was never about an ai jailbreak
level: technical
the us commerce department sent anthropic a letter on friday invoking an obscure export control directive. it banned non-americans, including anthropic employees, from accessing fable 5 and mythos 5. the letter cited an unspecified national security concern. anthropic believes it relates to a guardrail bypass but lacks details because the letter was not made public. in response, anthropic shut down both models for all customers to comply. the government forced the shutdown with a swift, unilateral action that did not require court approval.
new details cast doubt on the government's reasoning. cybersecurity researcher katie moussouris said anthropic shared a private paper describing an alleged guardrail bypass in fable 5. the paper's authors are reportedly security researchers at amazon. moussouris explained the bypass involved asking the model to fix code instead of reviewing it for security issues. she said the behavior cannot meaningfully be fixed and any attempt would weaken the model for defense. she and other experts called the export control directive hasty and misguided, urging the trump administration to revoke it.
the directive appears retaliatory given the administration's fractious relationship with anthropic. sources cited by axios said personality differences led to the action rather than a technical issue. the move sets a dangerous precedent for government control over american-made software. it also signals to foreign capitals that us ai companies may not be reliable for critical applications due to potential government interference. the administration has not confirmed its reasoning, leaving open questions about whether officials misread the report or acted out of spite.
why it matters: this shows how government actions can abruptly restrict access to ai models, impacting research and deployment without clear technical justification.
source: techcrunch ai: the us government’s anthropic models ban was never about an ai jailbreak