On June 30, 2026, the US Commerce Department lifted the emergency export controls on Anthropic’s Claude Fable 5 and Claude Mythos 5, ending an 18-day suspension that began on June 12. The restoration follows a comprehensive safety agreement between Anthropic and the US government.

The Agreement

The Commerce Department approved Anthropic’s safety package, noting in a statement that “Anthropic’s cooperative approach and swift action set a new benchmark for AI safety. The standards established here may serve as a template for other AI companies.”

Anthropic CEO Dario Amodei commented: “We learned a great deal from this experience. Safety and innovation are not in conflict—we have demonstrated they can go hand in hand.”

The Three-Pillar Safety Framework

1. Advanced Prompt Filtering

A multi-layered filtering system detects malicious prompts through both keyword-based filters and a semantic intent analysis engine powered by AI, blocking jailbreak attempts with high accuracy.

2. Inference-Time Guardrails

When the model attempts to output vulnerability code or dangerous instructions, the guardrails automatically intervene during the reasoning process, halting generation before harmful output is produced.

3. Regular Third-Party Security Audits

Independent security audits are now mandatory, with results reported to the Commerce Department. Non-compliance could trigger re-imposition of controls.

Service Restoration Conditions

Mythos 5 resumes service for US organizations with government approval, while Fable 5 is restored with full global access. Both models operate with the new safety measures in place.

Looking Ahead

This episode is expected to become a landmark case for AI governance worldwide. The rapid public-private collaboration demonstrated here offers a prototype for future AI regulation. The safety measures Anthropic implemented are likely to be adopted broadly as industry best practices.

From the June 9 launch through the June 12 controls and the June 30 restoration, the 18-day saga of Fable 5 and Mythos 5 will be remembered as a defining moment in AI history—one that proved safety and innovation can be achieved together.