The US government has lifted its export ban on Anthropic’s two most advanced AI models, Fable 5 and Mythos 5, just under three weeks after ordering them taken offline over national security concerns. Anthropic confirmed on the 1st of July that both models will be restored following a letter from Commerce Secretary Howard Lutnick confirming the restriction has been lifted.
For any business that noticed Claude behaving differently, or found certain model tiers unavailable, in mid-June: this is the explanation.
What caused the shutdown
Anthropic released Fable 5 and Mythos 5 on the 9th of June. Both were the most capable models the company had ever put in front of customers. Fable 5 was the consumer-facing version, a commercial release of what Anthropic calls its Mythos-class capability tier. Mythos 5, with fewer restrictions applied, remained available only to vetted organisations through Anthropic’s Project Glasswing programme.
Three days after launch, on the evening of the 12th of June, the US Commerce Department issued an emergency directive ordering Anthropic to cut off access to both models for any foreign national, anywhere in the world, including Anthropic’s own non-citizen employees.
The stated trigger was a jailbreak, a method of bypassing the model’s safety guardrails that the government claimed had been discovered and reported by a credible third party. Anthropic publicly disputed the severity of the finding. The company said the jailbreak it had reviewed was narrow, that similar techniques could be used against other publicly available models not subject to any export order, and that no universal bypass of Fable’s protections had been demonstrated.
Because Anthropic cannot verify the nationality of every user making an API request in real time, it could not comply with the directive selectively. The result was a complete, global shutdown of both models for every customer, regardless of location.
What was behind the government’s urgency
What was not in the public directive at the time of the shutdown has since become clearer. Reporting since the ban was issued indicates that Mythos, tested in a controlled exercise, was able to breach almost all of the US National Security Agency’s classified systems within hours. The NSA chief reportedly relayed this finding to the Senate. If accurate, it is the kind of capability demonstration that explains both the speed of the government’s response and the absence of a detailed public explanation in the directive itself.
The broader political context also matters. The Trump administration had previously ordered federal agencies to stop using Anthropic’s models following a contractual dispute with the Pentagon, and had designated Anthropic a “supply chain risk” for defence contractors. The export ban arrived against a backdrop of sustained friction between the company and the administration, which shaped how observers read the government’s stated rationale.
The terms for restoration
Access is being restored following Anthropic’s agreement to three commitments: proactively detecting and addressing security risks associated with both models; collaborating with the government on future model releases; and alerting authorities to any evidence of malicious activity. The Commerce Department has reserved the right to reimpose restrictions if it considers that necessary.
Anthropic has retained its position that the original directive was disproportionate. The company has not conceded that the jailbreak identified in June warranted a full global withdrawal, and it continues to challenge its Pentagon “supply chain risk” designation in federal court.
Dr Logic perspective
The most significant aspect of this episode for UK businesses is not the specific models involved. It is what the shutdown demonstrated about a category of risk that most organisations had not been accounting for: a commercially deployed AI tool can be removed from service by a foreign government, without a court order, without public documentation of the reason, and within hours.
For businesses using Claude, or any US-based AI platform, in operational workflows, this is a vendor continuity question as much as a technology one. The June outage was resolved without lasting disruption for most users, and access to other Anthropic models including Claude Sonnet was never affected. But teams that had embedded Fable 5 into production processes or automated workflows found themselves making emergency decisions about fallback options on a Friday evening.
The practical lesson Dr Logic would draw from this is the same one that applies to any critical tool dependency: know what your fallback is before you need it. That means understanding which of your workflows are tightly coupled to a specific AI model tier, and having an alternative path documented even if you never expect to use it.
For most businesses using Claude at a standard tier, day-to-day usage is unaffected, and the models are now fully restored. If your team is working with Mythos-class capabilities through Project Glasswing, you will be restored directly through Anthropic’s notification process.
For advice on building AI tools into your operations in a way that accounts for this kind of platform risk, speak to Dr Logic’s IT Strategy team.
