US Commerce lifted export controls on Claude Fable 5 and Mythos 5 on June 30, and Fable 5 returned to Claude.ai, Claude Platform, Claude Code, and Claude Cowork on July 1, ending an 18-day suspension following a June 12 Commerce directive. The controls were triggered by an Amazon researcher report describing a prompting technique that led Fable 5 to identify software vulnerabilities and produce exploit demonstration code. Anthropic trained a new classifier blocking that specific technique in over 99% of cases; blocked requests are rerouted to Opus 4.8 with a user notification, at the cost of more false positives on routine coding tasks. CAISI reviewed the safeguards before Commerce withdrew controls. Critically, Anthropic confirmed that multiple weaker models - including Opus 4.8, GPT-5.5, and Kimi K2.7 - could reproduce the same vulnerability-identification behavior, and every model tested could produce the same exploit demonstration code. For practitioners: a jailbreak Anthropic classifies as 'minor' (safety margin intrusion, not a core dangerous-capability breach) was sufficient to trigger policy action affecting all deployment channels.
AnthropicAI restores Fable 5, Mythos 5 access after US lifts export controls