US government forces global Fable 5 shutdown over a jailbreak GPT-5.5 can already replicate

The US government invoked export control to force Anthropic to disable Fable 5 and Mythos 5 for every customer worldwide — the cited jailbreak technique is asking the model to find software vulnerabilities, which GPT-5.5 already does without any bypass. This sets a precedent: export control authority can now compel a global AI model recall on verbal evidence of a capability publicly available in competing models.

Source: simonwillison.net ↗

These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.

Anthropic statement

Why this matters

→ Export control now forces global AI model recalls based on capabilities already public in competitors.
→ Sets precedent for government shutting down services via verbal evidence, no formal review process.
→ Jailbreak technique cited is routine vulnerability-finding that GPT-5.5 performs without restriction.

Export control overreach