US Forces Anthropic to Shut Down Claude Fable 5 and Mythos 5

Anthropic launched Claude Fable 5 on June 9, 2026, calling it the most capable model ever released to the general public. The model was a guardrailed sibling of Mythos 5, the cybersecurity-focused frontier model behind Project Glasswing, which had already identified over 10,000 high-severity vulnerabilities across critical software infrastructure.

Just three days later, on June 12, the US Commerce Department issued an export control directive ordering Anthropic to disable both models for all customers worldwide. According to reporting from Forbes and the BBC, the order was triggered by a reported jailbreak that demonstrated Fable 5 could be coerced into generating actionable exploit code, crossing a red line the government had drawn for dual-use AI capabilities.

Anthropic complied immediately, pulling API access and directing customers to Opus 4.8 as the recommended alternative. The shutdown affected both public API customers and the ~200 organizations participating in Project Glasswing, including ICE/NYSE and Rubrik, who had been using Mythos to self-vet critical infrastructure.

This is the first time a US government export control order has forced the shutdown of an already-deployed frontier AI model. It sets a precedent that will shape how AI labs time releases, structure safety evaluations, and coordinate with regulators. For developers building on frontier APIs, it's a stark reminder that model availability is not guaranteed, and production architectures need fallback strategies.