Anthropic's Fable 5 nears return after 18 days offline under a US export-control order

Fable 5, the model Anthropic billed as the most capable it had ever released to the public, has been dark since June 12. That afternoon the company got a letter from the US government, citing export-control authority, ordering it to cut off access to Fable 5 and its sibling Mythos 5 for any foreign national, inside the country or out, including Anthropic's own foreign-national employees. There is no clean way to enforce a rule like that on a hosted model, so Anthropic switched both off for everyone to stay compliant. Other Claude models stayed up.

Eighteen days later, the signals point to Fable coming back, maybe within the week. It is worth separating what is confirmed from what is still a guess.

What actually happened

The order traces back to a jailbreak. By Anthropic's account, the government believed someone had found a way around Fable's safeguards. What Anthropic says it was actually shown is narrow: a technique that "essentially consists of asking the model to read a specific codebase and fix any software flaws." The company reviewed the report it believes prompted the directive and said the capability on display is "widely available from other models," naming OpenAI's GPT-5.5, and is the same sort of thing defenders use every day to find and patch bugs.

According to Kalshi's reporting, which cites the Wall Street Journal, the trigger was Amazon. Researchers there bypassed some of Fable's guardrails, and Amazon CEO Andy Jassy reportedly phoned US officials to flag it after failing to reach Anthropic CEO Dario Amodei directly. (Amazon is a large Anthropic investor.) Amodei later argued, on calls with Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick, that Amazon's bypass was not a real jailbreak. Lutnick sent the export-control letter anyway.

The distinction Anthropic keeps pressing is between a non-universal jailbreak, which pries some restricted output loose under specific conditions, and a universal one, which broadly unlocks the model. The company says no one has found a universal jailbreak for Fable, that perfect resistance is not currently possible for any provider, and that it said as much at launch. Its position, stated plainly in the June 12 notice: "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."

Why people think it is coming back

The thaw is real and documented. On June 27 the Commerce Department let Anthropic restore Mythos 5, the stronger cyber-focused model that has never been openly available, for a limited set of trusted companies. Lutnick's letter said Anthropic "has worked with the U.S. government to address risks associated with" the two models and that "these efforts have yielded significant progress." Axios reported, citing two sources, that Fable access could be restored as soon as this week, with talks continuing through the weekend.

What is not settled is the sign-off. The Pentagon and the National Security Agency still have to clear Fable specifically. The Pentagon has been at odds with Anthropic for months, having labeled the company a "supply-chain risk to national security" in February over how Claude can be used in autonomous weapons. So a near-term return is a forecast, not a posted date.

The speculation, such as it is

Most of the loud public guessing is happening on prediction markets. As of June 17, Kalshi's market on when access returns priced a 57% chance before July 1, 67% before July 10, and 75% before July 17. By late June those numbers had drifted higher, with traders giving roughly a 74% chance of a return by mid-July. More than $450,000 has been staked on the contract, according to CNBC. That is a sentiment gauge. None of the people trading it have inside information.

There is also a question the markets do not answer, and it is the one most users care about. Anthropic had handed paid Claude subscribers free access to Fable through June 22 as a trial run. Axios notes it is not clear whether that free access comes back, or whether Fable returns behind extra fees or identity checks. The export order was written around foreign nationals, and verifying who is using the model is the obvious compliance lever, so some form of gating would not be a surprise.

The bigger fight

Both Anthropic and OpenAI are pushing the administration to replace one-off, case-by-case model reviews with a written process. Trump's June 2 executive order set up voluntary government vetting of the most capable models, with companies asked to share them up to 30 days ahead of wider release. Neither company wants access decided one phone call at a time. When OpenAI got a limited preview of GPT-5.6 cleared on the same Friday Mythos came back, it said the quiet part out loud: "We don't believe this kind of government access process should become the long-term default."

That is the actual story under the Fable blackout. A frontier model already in users' hands was pulled by government order over a disagreement about what counts as a dangerous capability, and no one has yet agreed on the rules for the next time it happens.