Anthropic restores Claude Fable 5 broadly, gates Mythos 5 to vetted partners after export freeze lifts

The restoration closes a roughly three-week saga that began when Amazon flagged a jailbreak in Claude Fable 5 capable of surfacing software vulnerabilities, prompting a US export suspension on June 12. On July 1 the Commerce Department reversed course, and Anthropic re-enabled Fable 5 across Claude.ai, Claude Code, the Claude Platform and Cowork, and returned it to Amazon Bedrock and Microsoft Foundry with 'global' and US data-zone options.

Mechanically, the fix is a dedicated safety classifier trained to detect and refuse the exact jailbreak pattern; Anthropic claims a >99% block rate. Crucially, Anthropic split its deployments: Fable 5 is the broadly available model, while Mythos 5 — which ships with fewer guardrails for offensive/defensive security work — is limited to trusted partners under Project Glasswing.

Competitively, the episode has become a flashpoint. Anthropic warned that applying the 'jailbreak' standard industry-wide 'would essentially halt all new deployments for frontier models,' a not-so-subtle jab at how a single reported vulnerability nearly froze a frontier release. The BleepingComputer report also clarified that Fable 5 is not permanently leaving consumer subscriptions, addressing panic among Pro/Max users.

What to watch: whether other vendors face the same classifier test, and whether the Mythos-for-partners-only model becomes a template for gating dual-use capabilities. Skeptics in the developer community suspect the speed of the reversal points to political or financial motives rather than a purely technical resolution.