Anthropic deploys Claude Sonnet 5, Fable and Mythos restored

Details

On July 1, Anthropic concluded an 18-day operational shutdown of its highest-capability frontier models — Fable 5 and Mythos 5 — triggered by a US government export control directive issued June 12. The pause followed Amazon researchers documenting a method to bypass Fable 5's safety controls, allowing the model to identify software vulnerabilities and generate exploitation code. During the shutdown, security evaluations found the same vulnerability replicable across models from multiple providers including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — suggesting the issue is systemic rather than unique to Anthropic. Engineers resolved the federal directive by training an automated safety classifier targeting the specific bypass mechanism. When triggered, it routes workloads to the older Opus 4.8 architecture to maintain service continuity, and internal validation data indicates it prevents the reported technique in more than 99% of trials. The fix carries a known trade-off: the classifier flags legitimate developer prompts more frequently during routine application development and debugging. Engineering teams transitioning agentic workloads are moving to the newly-launched Claude Sonnet 5, positioned as the cost-efficient backbone for autonomous agents. The episode marks the first government-mandated access blackout for a commercial frontier AI system, setting a significant regulatory precedent for the industry.