Anthropic’s Fable 5 Export Ban Lifted, Setting a Bar for Every AI Lab

The US Department of Commerce lifted export controls on Anthropic’s Claude Fable 5 and Mythos 5 AI models on Tuesday, less than three weeks after the company was ordered to suspend access to both models on June 12 over national security concerns about a “jailbreak” of Fable 5. Anthropic said it would begin restoring access on Wednesday. The commitments Anthropic made to Commerce to lift the ban now set the bar for every other frontier AI lab waiting in the government’s licensing queue.

Commerce Secretary Howard Lutnick confirmed in a letter to Anthropic, seen by the BBC, that “a license is no longer required for the export, reexport, or in-country transfer, including deemed export or deemed reexport, of the Mythos or Fable models.” Anthropic posted on X on Tuesday that it would share further updates as restoration proceeds and thanked its users for their patience.

From Launch to Global Lockout

Anthropic released Fable 5 and Mythos 5 on June 9, 2026, calling them its most advanced reasoning and cybersecurity models. Three days later, at 5:21 p.m. ET on June 12, the Commerce Department’s Bureau of Industry and Security sent CEO Dario Amodei an export-control directive signed by Lutnick.

The letter ordered Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including the company’s own non-citizen employees. Because Anthropic said it could not reliably tell its users apart by citizenship in real time, it disabled both models for everyone worldwide. Other Anthropic models, including Claude Opus 4.8, remained available. The disruption hit developers, researchers, and enterprise customers globally, per Anthropic’s June 12 statement on the directive.

1. June 9, 2026: Anthropic releases Fable 5 and Mythos 5
2. June 12, 2026, 5:21 p.m. ET: Commerce directive received, signed by Lutnick
3. June 14, 2026: Open letter at freefable.org from information-security executives
4. Last Friday: Partial restoration of Mythos 5 to a small group of trusted US cyber defenders
5. Tuesday, June 30: Full lifting announced; Wednesday, July 1: global restoration begins

What Anthropic Conceded to Commerce

The lifting did not come with a clean reversal. It came with three explicit commitments in a letter from Lutnick, seen by the BBC and Reuters. Anthropic agreed to proactively detect and address security risks associated with the models; to work with the US government on protocols, standards, and releases for Mythos, Fable, and future models; and to inform the government of any malicious activity involving the systems.

• Proactively detect and address security risks associated with the models
• Work with the US government on protocols and standards for Mythos, Fable, and future model releases
• Inform the US government of any malicious activity involving the systems

The third commitment is the one with the longest reach. By pledging to collaborate on standards for “future models,” Anthropic volunteered to help write the rulebook Commerce will apply to its competitors. Lutnick said on X that the Bureau of Industry and Security had withdrawn the controls after evaluating “the diversion risks now presented” by the two models, and that the agency’s decision could be revisited. The Department of Commerce did not respond to questions from reporters about which specific commitments triggered the lift.

The deal’s effective window is also unbounded. The Commerce Department “reserves the right to reconsider its decision to lift export restrictions if necessary,” per the Lutnick letter. Anthropic said in its Tuesday update on the lifting that it was “grateful to our users for their patience, and to everyone who worked with us on redeploying the models.” The company is also working with Amazon, Microsoft, Google, and other Glasswing partners on a proposed industry framework for scoring jailbreak severity, per a separate post on Anthropic’s site titled “Redeploying Fable 5.”

The Frontier Lab That’s Still Locked Out

OpenAI’s GPT-5.6 has not cleared the same bar. Reuters reported on the same day that OpenAI delayed a full public launch of GPT-5.6 at the US government’s request, keeping the model available only to a small group of vetted partners. There was no public timeline for a wider release, and an OpenAI spokesperson did not respond when POLITICO asked whether the company had heard anything from the White House about when the hold would end.

Industry advisers read the Anthropic deal as a one-off, not a template. Joe Hoefer, chief AI officer at Monument Advocacy, told POLITICO the arrangement “makes sense given how new this territory is” but warned it was “a case-specific agreement, not a standard the next company can point to.” One industry observer quoted in the same story put it more bluntly: “We have no idea what Anthropic did to make the models ‘safe,’ what commitments Anthropic has made going forward, and whether or how any of this applies to other frontier models in the government’s licensing queue.”

Extensive safety testing is not a bad idea. I just don’t like the idea of the government picking the customers.

OpenAI CEO Sam Altman pushed back on the government’s role in selecting which companies can access frontier AI, writing on X last week that extensive safety testing “is not a bad idea” but objecting to Washington handpicking customers. OpenAI’s still-locked GPT-5.6 is the next live test of whether Anthropic’s deal holds as a precedent or is a one-off.

Two Stories About the Same Jailbreak

Anthropic and the administration have given two sharply different accounts of what triggered the original directive. In its June 12 statement, Anthropic said the government’s concern was a method of bypassing, or “jailbreaking,” Fable 5’s safeguards against offensive cyber use. The company reviewed the demonstration at the heart of the directive and concluded that the bypass was already a standard feature in rival systems, including OpenAI’s GPT-5.5. Anthropic called the technique “narrow” and “non-universal”, and said no tester had found a universal jailbreak on Fable 5 despite thousands of hours of pre-launch red-teaming.

We have reviewed a report that we believe is the basis of the government’s directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.

The administration’s most visible public voice on the matter has been White House AI and crypto adviser David Sacks, who characterized the same bypass as a serious exposure of a cyber-capable model and said the company had declined a reasonable request to fix or withdraw it before the control was imposed. The Commerce Department itself declined to comment when reporters asked for an on-record rationale, so the government’s full reasoning has not been published.

The technical question sits inside a bigger argument. On June 14, an open letter at freefable.org signed by information-security executives called for the directive to be lifted and for any future AI risk regulation to be “scientific, democratic, transparent, and applied only to the minimal extent necessary.” Anthropic went further: if the standard applied to Fable 5 “was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” the company said in its June 12 statement.

How Access Returns, and What the Industry Inherits

Anthropic said it would begin restoring access to both models on Wednesday, July 1, with broader availability rolling out over the days that follow. A partial easing came last Friday, when the government allowed Anthropic to provide Mythos 5 to a small group of trusted US organizations, including cyber defenders. Fable 5 stayed locked through that partial step and reopens Wednesday alongside the wider release, with the earlier partial Mythos 5 approval now fully in the rearview.

The lifting left several operational questions open. The directive was withdrawn, but the broader “deemed export” framework that made it possible is still in force; foreign nationals inside the US may still face access limits on future frontier models, depending on how Anthropic implements the new monitoring and reporting commitments. Anthropic said it would share further updates on access as restoration proceeds.

The bigger question hangs over the rest of the industry’s launch calendar. Glenn Gerstell, the former general counsel of the National Security Agency from 2015 to 2020, called rescinding the export restrictions “the right step” but said the episode “underscores how we have a lot more work to do to figure out the right way of balancing regulation for safety and national security reasons with promoting innovation in the AI sector.” Hoefer, at Monument Advocacy, said the current outcome amounted to “more of a ceasefire” than a real resolution: “Until there’s a codified process with defined triggers and a real evidentiary standard, every frontier model launch carries the risk of a repeat.”