Claude Fable 5, Anthropic’s most capable publicly available model, returns to global service on July 1 after a 19-day suspension under a US Department of Commerce export control directive.

What happened

Fable 5 launched on June 9, 2026. Three days later, the US government ordered access suspended after Amazon researchers reported a technique capable of bypassing the model’s safety guardrails — prompting it to identify software vulnerabilities and demonstrate their exploitation. Anthropic and the US government worked through a 19-day review while access was cut to zero for all users worldwide.

How access is being restored

The Trump administration lifted the export controls on June 30. Anthropic has deployed a new safety classifier targeting the reported bypass technique, which blocks it in more than 99% of cases. The company also noted that testing found many less capable models could produce comparable vulnerability information, suggesting the suspension’s marginal safety benefit was limited.

Fable 5 will be available across Claude Platform, Claude.ai, Claude Code, and Claude Cowork starting July 1.

Industry jailbreak severity framework

Alongside the redeployment, Anthropic has proposed an industry-wide standard for scoring jailbreak severity, developed with Amazon, Microsoft, Google, and other Project Glasswing partners — as well as US government agencies including the Office of National Cyber Director and the Department of Commerce’s CAISI. The framework defines four criteria:

  • Capability gain — how far the jailbreak extends beyond what existing tools already offer
  • Breadth of capability gain — how many distinct offensive tasks it enables
  • Ease of weaponization — how much human effort is required to exploit it
  • Discoverability — how accessible the technique is to bad actors

The framework is designed to give the industry and regulators a shared vocabulary for assessing jailbreak severity, rather than treating all safety bypasses as equally serious.

Why it matters

The Fable 5 suspension was the most disruptive AI export control action to date, creating gaps in enterprise deployments and raising questions about how governments would handle future safety incidents. The redeployment, paired with a multi-company severity framework, marks the first concrete attempt to establish industry norms for what constitutes a dangerous bypass — and at what point government intervention is warranted.