The suspension lasted less than a month. On Wednesday, Anthropic announced it would begin restoring global access to Claude Fable 5 and Mythos 5 after the US Department of Commerce notified the company that export restrictions on both models had been lifted.

The two platforms — among the most capable AI tools commercially available — had been abruptly suspended on 12 June, just three days after their release, over concerns that they could be weaponised by hackers to exploit vulnerabilities in computer systems.

A Narrow Jailbreak, a Global Suspension

At the time of the suspension, Anthropic pushed back with unusual candour. The company said US authorities had not identified specific, concrete concerns about its technology, but had flagged what it described as a "narrow potential jailbreak" in Fable 5 — a technique that could allow users to slip past the model's built-in safety restrictions and unlock blocked capabilities.

Anthropic did not accept that finding as proportionate justification for recalling a product already deployed to hundreds of millions of users worldwide.

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company said at the time.

The tension between Anthropic's commercial position and Washington's national security calculus was visible from the start. The government moved quickly and broadly; Anthropic protested, but complied.

The Terms of Reinstatement

The reversal came after Anthropic entered into a formal agreement with the Commerce Department. In a letter reviewed by the BBC, Commerce Secretary Howard Lutnick confirmed the terms: Anthropic has agreed to proactively detect and address security risks associated with both models, collaborate with the government on future AI releases, and alert authorities to any malicious activity identified on its platforms.

Lutnick also made clear the arrangement is conditional — the Commerce Department reserves the right to reimpose restrictions if circumstances change.

The agreement represents something of a template. As AI capabilities expand — particularly in domains like cybersecurity, where tools such as Mythos 5 are explicitly designed to identify and exploit weaknesses in computer code — the question of who controls access to these systems, and on what terms, is becoming structurally important.

What the Two Models Do

Understanding why Washington moved so fast requires understanding what these tools actually are.

Fable 5 is Anthropic's consumer-facing model. Built on the Claude platform — a direct competitor to OpenAI's ChatGPT and Google's Gemini — it is capable of deep reasoning and can independently perform complex, multi-step tasks. It was designed for the mass market and had already achieved wide distribution before the suspension.

Mythos 5, by contrast, is built for businesses and cybersecurity professionals. Its core capability — identifying vulnerabilities in computer code and, critically, exploiting them — is precisely what made regulators nervous. In the wrong hands, or accessed via a compromised safety layer, such a tool carries obvious dual-use risk.

Both were released on 9 June. The suspension came three days later. The lifting of restrictions arrived weeks after that, contingent on Anthropic accepting an ongoing oversight relationship with the federal government.

A New Kind of Regulatory Relationship

What has quietly emerged from this episode is less a one-time intervention and more an evolving framework. Anthropic is now formally embedded in a government notification and collaboration structure — one that gives the Commerce Department early visibility into future model releases and a standing mechanism to act if security concerns resurface.

For an industry that has operated at arm's length from regulators for most of its short history, that shift is significant. The speed of the original suspension — and the fact that it was accepted, however reluctantly, by one of Silicon Valley's most credible AI safety organisations — signals that the US government is prepared to use export control architecture as a tool of AI governance, not merely as a legacy instrument for managing hardware or traditional software.

Whether that framework is the right one for managing the risks of frontier AI remains contested. What is no longer contested is that Washington intends to use it.