Commerce Department Lifts Export Controls on Anthropic Fable 5 and Mythos 5 After Jailbreak Review

Anthropic disabled worldwide access to Fable 5 in early June after the Commerce Department's Bureau of Industry and Security applied novel export controls to AI weights rather than silicon. The move followed an internal assessment of a multi-step prompt chain that fed known-vulnerable open-source code plus planted flaws into the model and manually assembled working patch-test scripts. Katie Moussouris's independent review for Anthropic concluded the chain was standard defensive find-fix-test workflow, not a novel guardrail bypass, and that equivalent outputs could be obtained from non-frontier models without comparable controls.

The reversal required Anthropic to deploy a new safety classifier blocking the technique in over 99 percent of cases, validated by the Commerce Center for AI Standards and Innovation, plus pre-release evaluator access, rapid jailbreak disclosure, and a HackerOne program. Mythos 5 remains gated to U.S. vetted entities under Project Glasswing while broader domestic and allied access is negotiated. Five Eyes statements simultaneously warned that frontier models will transform offensive and defensive cyber operations within months, not years.

An open letter signed by more than 100 practitioners, including Alex Stamos, warned that blanket export controls would degrade defensive tooling available to allied defenders while Chinese labs continued unrestricted training. The episode reveals selective enforcement: the same capability surfaced in OpenAI and Moonshot models yet triggered no comparable action, exposing the gap between stated national-security rationales and commercial competitive effects.

Next steps include an industry scoring framework across capability gain, task breadth, weaponization ease, and discoverability, plus expanded Glasswing participation by Amazon, Microsoft, and Google. The precedent sets the procedural template for future frontier-model controls and prefigures tighter coupling between U.S. evaluators and commercial labs.