Claude Fable 5 and Mythos 5 Restored After 19-Day Government Shutdown
View original source →On July 1, 2026, Anthropic restored global access to Claude Fable 5 and Mythos 5 after the US Commerce Department lifted its emergency export control order on June 30 — ending a 19-day suspension that disrupted enterprise AI workflows worldwide and established a landmark precedent for how governments can intervene in commercial AI deployment.
The shutdown began on June 12 when researchers identified a specific prompt sequence that could bypass Fable 5's safety filters and reveal software vulnerabilities. The US government applied emergency export controls, citing concerns the capability could be used to attack critical infrastructure or leak techniques to foreign adversaries.
Key points:
• The restoration required Anthropic to implement a new AI safety classifier that detects and blocks the specific jailbreak technique in over 99% of attempts
• Enterprise users accessing Fable 5 through Amazon Bedrock, Google Cloud, and Microsoft Foundry must now complete nationality-based identity verification
• Anthropic agreed to a consensus industry framework developed jointly with Amazon, Google, and Microsoft for measuring and communicating safety bypass severity to governments
• Mythos 5 — the unrestricted cybersecurity-focused model — remains gated, available only to approved domestic organizations in the Project Glasswing critical infrastructure consortium
• Usage is temporarily capped at 50% of weekly subscription limits for Pro, Max, and Team users through July 7
The 19-day shutdown accelerated enterprise trials of foreign AI alternatives. Zhipu AI's GLM-5.2 model reportedly matched Mythos on standard cybersecurity benchmarks during the shutdown period — illustrating that aggressive domestic bans can inadvertently drive buyers toward foreign models.
Why It Matters: This is the first time the US government has used export control laws to suspend a commercially deployed software model mid-market. The resolution establishes a new playbook: voluntary safety commitments are no longer sufficient — labs must demonstrate measurable technical fixes and participate in industry-wide safety standards before restoration is granted.