US Hits Anthropic With Export Order on Mythos 5 and Fable 5
2 min readThe US government has ordered Anthropic to cut global access to its two newest frontier models, Mythos 5 and Fable 5, citing alleged jailbreak risks. Anthropic publicly disputes the directive, calling it a “misunderstanding” and saying it is racing to restore service.
The order lands less than a week after Anthropic launched Claude Fable 5, its first model in the Mythos class, alongside an updated Mythos 5. Both sit at the top of Anthropic’s lineup and are now reportedly subject to export control restrictions that pulled global API access offline shortly after the request landed.
What Washington is alleging
According to news roundups circulating this week, the directive frames Mythos 5 and Fable 5 as dual-use technologies under existing export control rules, with regulators pointing to recent third-party jailbreak demos as evidence that safety mitigations can be peeled back faster than Anthropic claims. The order is global in scope, which means non-US customers lose access too, not just sanctioned regions.
Anthropic’s response has been unusually direct. The company says regulators have misread the model card, conflated red-team scenarios with real-world capability, and applied a control regime designed for chips to a hosted API. Anthropic is “working to restore access” and has framed compliance as a temporary stop-gap while it appeals.
A second front opens at the Pentagon
The export fight follows a separate Pentagon disclosure that it is actively testing OpenAI and Google models as potential replacements for Claude in classified systems. Anthropic has been the incumbent provider for sensitive defense workloads since late 2025, and a swap would cost it one of its most visible enterprise wins. Taken together, the two stories paint the same picture: a US government that is suddenly less comfortable with its biggest domestic frontier lab.
Why it matters
For customers, the immediate question is whether production traffic on Mythos 5 or Fable 5 needs a fallback plan today. For Anthropic, the bigger question is precedent. If a frontier model can be pulled offline globally on a jailbreak-risk theory, every lab now has a new compliance vector to plan around. Watch for Anthropic’s formal appeal, any public guidance from the Bureau of Industry and Security, and whether Microsoft Foundry and AWS Bedrock keep their Mythos-class deployments live during the dispute.