June 12, 5:21 PM ET: The Commerce Department sent a letter to Anthropic, ordering an immediate shutdown of Fable 5 and Mythos 5 access for all non-US citizens — including the company‘s own foreign-born employees. The models went offline. No appeal. No transition period. Just a letter and a deadline.
June 26: Mythos 5 was restored for more than 100 approved U.S. organizations for defensive cybersecurity purposes.
July 1: Fable 5 is expected to be restored. The full shutdown lasted 19 days.
The speed of the reversal is notable. Export controls are rarely reversed this quickly. The administration’s willingness to walk back its own order suggests the initial decision may have been more aggressive than intended — or that the political and economic pressure to restore access was greater than expected. Either way, 19 days is a short window for a policy that was presented as a national security necessity.
From Export Control to Licensable Framework
The administration is moving from an export control order to what one official described to Axios as an arrangement of “conditions and oversight.” The exact terms are still unclear, but the shift signals a new approach to regulating frontier AI models: not a permanent ban, but a licensable framework.
Anthropic CEO Dario Amodei was sidelined from negotiations after government officials found him difficult to work with. Co-founder Tom Brown took over — and progress followed. Anthropic will likely need to maintain ongoing security protocols and demonstrate compliance with U.S. export laws to keep access open.
The government has also announced plans to establish a formal process for licensing future AI model exports, with specific guidelines expected to be finalized by the end of summer. That means Fable 5 isn‘t just returning — it’s returning as the first model subject to a new regulatory regime.
The implications are significant. A licensing framework is not the same as a blanket ban. But it is also not the same as a free market. Under this new regime, Anthropic will need to demonstrate ongoing compliance with U.S. export laws to keep Fable 5 available. The company is also required to maintain government-approved security safeguards, including a user access system that can be audited by U.S. officials. This moves frontier AI models from the domain of commercial products to the domain of regulated technologies — closer to how the U.S. treats advanced aerospace or nuclear technology.
What Happened to the “Voluntary” Framework?
The Fable 5 episode has also raised questions about the viability of the administration‘s voluntary AI review framework.
When President Trump signed the AI executive order on June 2, it established a voluntary 30-day pre-release review window and explicitly disclaimed any “mandatory governmental licensing, preclearance, or permitting requirement.” Industry lawyers warned at the time that “voluntary today” could become “baseline tomorrow.” That warning is now reality.
The administration has effectively established a de facto licensing regime — not through legislation, not through formal rulemaking, but through a series of case-by-case decisions. Anthropic was the first test. OpenAI’s GPT-5.6 was the second. The framework is being built case by case, not policy by policy.
The White House hasn‘t said when the formal licensing guidelines will be completed, but officials have indicated they expect to finalize the process by the end of summer. That timeline suggests the administration plans to lock in the new rules before the end of the year — and before any potential leadership change could reverse course.
The Business Implications
For Anthropic, the restoration is a clear win. Fable 5 is its flagship commercial model, and a 19-day blackout was costly — not just in lost API revenue, but in enterprise credibility. Customers who rely on Fable 5 for critical workflows need confidence that the model will stay online.
But the return comes with conditions. Anthropic will need to maintain ongoing compliance with U.S. export laws, keep government-approved security safeguards in place, and develop a user access system that can be audited by U.S. officials at any time. That’s a new operational burden.
The regulatory uncertainty is now a permanent feature of the AI landscape — not a one-time event. For Anthropic, that means IPO timing matters. A licensing framework is easier for investors to model than a sudden shutdown. But it also means the company’s ability to control its own product distribution is now shared with the U.S. government.
What This Actually Means
The Fable 5 shutdown was a shock to the industry — the first time the U.S. government had used export controls to pull a commercial AI model offline. The restoration is a signal that the administration does not intend to make this a permanent policy, but a test case for a new regulatory framework.
The government has also announced plans to establish a formal process for licensing future AI model exports, with specific guidelines expected to be finalized by the end of summer.
The next test case is already visible: GPT-5.6 Sol was released in late June under a similar approval framework, and the White House is likely to treat future Anthropic and OpenAI model releases similarly. The rules of the road are being written in real time, one model at a time.
P.S. Fable 5 was offline for 19 days. That‘s shorter than most people expected — but longer than Anthropic wanted. The real story isn’t the shutdown or the restoration. It‘s what happens next. The administration is building a licensing system for AI models, and Fable 5 was the first test case. It won’t be the last. When GPT-5.7 ships, it will ship under the same rules — and by then, the government will have a playbook.
