A three-part series on the rise, sudden removal, and contested aftermath of Claude Fable 5. Part 2 covers the export-control order that pulled Fable 5 and Mythos 5 offline three days after launch.

Fable 5 had been available to the public for about 72 hours when the order arrived. Anthropic says the U.S. government’s directive reached the company on Friday, June 12, at 5:21 p.m. Eastern. By midnight its two most powerful models were gone, not slowed down and not limited to certain regions, but shut off for every user in the world.

The order

The legal instrument was an export-control directive. Commerce Secretary Howard Lutnick sent a letter to chief executive Dario Amodei, first reported by Axios and later obtained in full by Bloomberg, placing Claude Fable 5 and Claude Mythos 5 under export controls and requiring a license for any export, re-export, or domestic transfer of the models. The decisive part was the scope. The order barred access by any foreign national anywhere on earth, including foreign-national employees working inside Anthropic itself, and it carried explicit penalties for noncompliance. Anthropic said the letter offered no specific account of the national-security concern driving it.

The reporting filled in some of what the letter did not. An administration official told Axios that Commerce moved after another company claimed it had found a way to jailbreak Mythos, and that the order followed an earlier, failed attempt to get Anthropic to delay the launch on its own. The same official framed the lockdown as temporary, saying the models would need to stay offline until the government had hardened its own security posture, possibly within weeks. Several outlets reported that Lutnick’s letter raised the risk of the models being used by military-intelligence services in countries of concern, naming China and Russia.

Why everyone lost access

Read literally, the order did not shut Fable down for Americans. It restricted foreign-national access. The blanket worldwide blackout was a byproduct of how this kind of control actually works. Under U.S. rules, letting a foreign national use a controlled item on American soil counts as a so-called deemed export, so the restriction reached not only users abroad but, for example, a foreign-born engineer sitting in Anthropic’s San Francisco office. There is no dependable way to verify the nationality of every person using a cloud service in real time, and certainly not on the day a letter lands, across a user base in the hundreds of millions. Faced with a rule it could not enforce selectively, Anthropic concluded that the only way to comply was to turn the models off for everyone. Every other Claude model, including Opus, Sonnet, Haiku, and the older releases, stayed up and untouched.

The cutoff was sudden enough that developers watched it happen live. One well-known engineer ran a script that pinged the model on a loop and recorded the moment his access died that evening, when the calls started coming back with an error telling him the model was unavailable and that he should use Opus 4.8 instead. Anthropic had been giving paying subscribers free access to Fable through June 22, so the people most exposed were often the ones who had moved fastest. Some of them were in the middle of projects that suddenly had no model underneath them.

Anthropic’s response

The company did something a business on the receiving end of a federal order rarely does. It complied loudly and disagreed just as loudly. In a statement posted that night, Anthropic said it was removing access to both models in order to satisfy the directive, called the action a misunderstanding, and said it was working to restore access as quickly as possible.

It also laid out its read of the underlying problem. The company described the issue as a narrow, non-universal jailbreak, essentially a technique that involved asking the model to read a codebase and identify software flaws. It said it had reviewed what it believed was the report behind the order and concluded that the capability on display was available from other publicly deployed models, including OpenAI’s GPT-5.5. None of its own testers, it added, had found a universal jailbreak, and it had not received any disclosure of a non-universal one that produced a genuinely harmful result. Its central objection was about proportion. Recalling a commercial model used by hundreds of millions of people over a narrow potential jailbreak was, in the company’s view, an overreaction, and a standard that, applied across the industry, would effectively stop every frontier lab from shipping anything new. Anthropic promised more detail within 24 hours.

The immediate cost

The practical damage was instant and uneven. Developers who had spent launch week building on Fable were cut off without warning, their tooling replaced by an error message. Security teams that had started using the model for legitimate defensive work lost it overnight. The 30-day data-retention requirement that Anthropic had attached to its Mythos-class models, already a sore point for corporate customers who had been promised their data would not be kept, now sat alongside a sudden, total outage of the product those customers had just adopted.

For Anthropic the timing was its own kind of injury. This was the launch that was supposed to prove the company could bring frontier capability to the public without setting off a disaster. Three days in, it had been undone by precisely the sort of government intervention its own chief executive had argued for the day after launch. What had looked on Tuesday like the company’s biggest release of the year had become, by the weekend, a fight it could not afford to lose in public.

Next, Part 3: the competing explanations, the three-word “jailbreak,” the Korean telecom nobody saw coming, and where the standoff stands now.


Sourcing: Anthropic’s public statement of June 12; reporting from Axios, Bloomberg, the Wall Street Journal and Quartz; and contemporaneous developer accounts of the cutoff. The full text of the Lutnick letter has been reported but not officially released by the Commerce Department, so the rationale attributed to the order reflects reporting and Anthropic’s own characterization.