Anthropic pulls Mythos 5 and Fable 5 after US export control threat

Anthropic has pulled access to Mythos 5 and Fable 5 after the US government warned it could impose export controls unless the company restricted the models, according to reporting from The Verge published on June 16. The move took two of Anthropic’s newest cybersecurity-focused systems offline just days after launch and turned what looked like a product rollout into a live test of how far Washington is willing to reach into frontier AI releases.

This matters well beyond Anthropic. If US officials can force one lab to disable a flagship model on short notice, the precedent reaches OpenAI, Google, Microsoft, and any startup selling advanced AI systems globally. The immediate question is whether governments should be able to freeze access this quickly. The longer-term question is whether model launches now need the same political-risk planning as chip exports or defense contracts.

What triggered the shutdown

The Verge reports that Anthropic received a Friday directive tied to export-control authority and was told to suspend access for foreign nationals, including foreign-national employees. Anthropic said it believes the order stemmed from concerns about a bypass of Fable 5’s safeguards, but argued the issue was narrow rather than a universal jailbreak. TechCrunch, citing people familiar with the situation and outside security experts, separately reported that the government letter did not publicly spell out the underlying national-security concern.

That detail matters because Mythos 5 and Fable 5 were positioned as high-end cybersecurity models. Supporters say systems like these can help defenders find and patch vulnerabilities faster. Critics worry the same capability can be turned toward offensive misuse. That dual-use tension has existed for years in security tooling, but AI models compress it into a far more scalable product.

Why the industry is watching

Anthropic’s scramble to negotiate with the Trump administration shows that frontier AI policy is no longer limited to voluntary commitments and safety blog posts. According to The Verge, executives and researchers spent the weekend in talks with US officials while outside security leaders pushed for the restrictions to be reversed. If comparable models from rival labs offer similar capabilities, a targeted action against Anthropic could eventually become a broader template for regulating frontier cybersecurity AI.

For now, the biggest takeaway is not whether Mythos 5 returns this week or next. It is that frontier model access can now be interrupted by export-control pressure even without a long public rulemaking process. That raises the cost of launching advanced AI systems, strengthens the hand of regulators, and gives enterprise buyers one more risk factor to consider before building security workflows around a single vendor.