Both models are gone. Not degraded, not rate-limited. Fully offline, globally, as of June 12, because the U.S. government said so.
According to Anthropic’s official statement, the U.S. government issued an export control directive prohibiting any foreign national from accessing Claude Fable 5 or Claude Mythos 5. That restriction sounds targeted. It isn’t. Because Anthropic can’t segment foreign nationals from domestic users at the model serving layer without shutting the whole system down, the directive’s practical effect was a complete global suspension. Every customer, U.S. or otherwise, lost access on June 12, 2026.
The stated basis was a jailbreak vulnerability. The government cited a jailbreak method capable of bypassing the models’ safeguards in ways that could assist malicious actors, according to the Associated Press. Anthropic disputed that framing. The company characterized the issue as involving only a small number of previously known, minor vulnerabilities, per its own statement, a vendor claim, not an independently confirmed assessment. The underlying technical details remain classified or undisclosed. There’s no public record of exactly what the jailbreak does or how severe the government’s assessment actually was.
The catch is that it doesn’t matter whether Anthropic was right. Under export control law, compliance isn’t optional. When the directive came, Anthropic had one choice: shut the models down.
Fable 5 Suspension: Party Positions
Claude Opus and the rest of Anthropic’s model lineup weren’t named in the directive. They remain available. The suspension is limited to Fable 5 and Mythos 5, the two models Anthropic had positioned as its most capable frontier releases, with Anthropic’s internal evaluations citing significant performance advances. Independent benchmark evaluation by Epoch AI had not been completed at the time of the suspension, so any specific benchmark figures for either model remain at the vendor-reported tier.
What this means for teams in production
If you’re running Fable 5 or Mythos 5 in any production workflow, agents, RAG pipelines, API integrations, those systems are down. The immediate fallback is Claude Opus. Whether Opus performs comparably for your specific workload depends on the task; the benchmark gap between Opus and the suspended models, per Anthropic’s own evaluations, was material. Don’t expect a seamless swap.
This didn’t happen in a vacuum. The directive follows the June 2, 2026 Trump executive order establishing a voluntary 30-day vetting framework for advanced AI systems. That framework was voluntary. Ten days later, the government used a different and significantly harder instrument, an emergency export control directive, to force a shutdown. The escalation from voluntary framework to emergency directive is the part nobody mentions in the headlines, but it’s the signal that matters for enterprise planning.
What to Watch
Duration of the suspension is unknown. Anthropic hasn’t indicated a timeline. The government hasn’t specified conditions for restoration.
TJS synthesis
Export controls aren’t new. Applying them to AI models, at the model API layer, is. If you’re building production systems on frontier models, add a new failure mode to your incident response planning: not vendor outage, not deprecation, but government-ordered suspension with no advance notice and no stated end date. Build your model governance framework to account for it. That means multi-vendor fallback architecture, not just multi-model. See prior TJS coverage on government-Anthropic security risk framing for context on how this relationship has developed. The June 12 directive is a new escalation in that arc. Wait for Anthropic’s full statement to be publicly accessible before making any final architectural decisions, but don’t wait to start the contingency conversation.