Fable Remains in Limbo, Restoration Probabilities Rise

Anthropic's Fable 5 and Mythos 5 models have been fully suspended since June 12, 2026, when the US government issued an export control directive blocking all access for foreign nationals, whether inside or outside the United States. Anthropic received the directive at 5:21pm ET; to comply, it disabled both models for all customers globally, including foreign national Anthropic employees. Access to all other Anthropic models was unaffected.

Anthropic stated the only demonstrated jailbreak was a narrow, non-universal technique involving asking Fable 5 to read a specific codebase and identify software flaws. Per Anthropic's official statement, the vulnerability "is widely available from other models (including OpenAI's GPT-5.5), and is used every day by the defenders who keep systems safe." Anthropic had invested thousands of hours of red-teaming before launch with the US government, UK AISI, and private third parties, and said no tester had found a universal jailbreak. The company publicly called the suspension a likely misunderstanding and stated it disagrees that a narrow jailbreak finding should warrant recalling a commercial model deployed to hundreds of millions of users.

Zvi Mowshowitz's June 25 AI roundup (AI #174) gives 45% odds that access resumes by June 26, and 69% by July 1 - his own probability estimates based on publicly available information. Earlier, on June 17, Anthropic's Managing Director of International Chris Ciauri said Anthropic is "very confident that in the coming days, the models will become available again."

The now-public capabilities post describes Fable 5 as exceeding all prior Anthropic releases. Benchmarks include the highest score on FrontierBench (Cognition's frontier coding eval), the first model to break 90% on Anthropic's core analytics benchmark (a 10-point gain over Opus 4.8), and near-parity with GPT-5.5 on frontier physics research using a third of the reasoning tokens. Safety safeguards routed less than 5% of sessions to the fallback Opus 4.8 model; Anthropic described these as intentionally conservative and noted they generated complaints of over-restriction.

The suspension marks one of the first instances of Washington applying export controls to a deployed commercial AI model rather than to hardware or chips, setting a precedent with broad implications for how frontier labs ship products globally, which customers they can serve, and where their engineers can be based.