Anthropic says the US Department of Commerce has removed export controls on Claude Fable 5 and Mythos 5, clearing the company to restart its consumer-facing Fable model after weeks of talks with the Trump administration.
In a post on X, Anthropic said it would start bringing Fable 5 back to Claude users around the world on Wednesday. The company said access through AWS, Google Cloud, and Microsoft Foundry would follow, though it did not give a date. For customers who built around the model during Anthropic’s launch push, that means the outage is ending first on Anthropic’s own Claude platforms, with the cloud reseller plumbing still waiting in the queue.
The fight began in early June, when Anthropic pulled Fable 5 offline after the Trump administration issued an export-control directive. According to The Verge, the order barred foreign nationals from using Fable 5 or Mythos 5, including non-US staff at customer companies and many Anthropic employees. Fable 5 is Anthropic’s consumer model based on the same underlying technology as Mythos 5, with additional safeguards layered on top.
The jailbreak fix Anthropic is selling
Anthropic said the export restriction was tied to concerns about jailbreaks, meaning prompts or workflows that can push a model past its safety rules. The company said Amazon researchers had identified the technique that helped trigger the government action.
In a blog post, Anthropic said it trained a new safety classifier aimed at detecting and blocking that behavior. When the classifier blocks a Fable 5 request, Anthropic said the user will be told, and the request will be routed to Opus 4.8 instead. Anthropic claims the classifier stops the Amazon-described technique in more than 99 percent of cases.
That is Anthropic’s claim, not proof that Fable 5 is jailbreak-proof. The company also said no universal jailbreak for Fable 5 had been found at the time of writing, while acknowledging that full robustness against jailbreaks is probably not possible. That is a refreshingly non-magical sentence for an AI lab, even if it arrives after a regulator-sized crater opened under the product rollout.
More government access before release
Anthropic also laid out a more formal relationship with the Trump administration for future model launches. The company said it plans to give government partners pre-release access and evaluation windows, especially for models with national security relevance. During those tests, government evaluators would be able to examine model capabilities and guardrails, with access to Anthropic technical staff.
Anthropic said it will also set up faster information sharing when major jailbreaks or misuse patterns appear. The company said it intends to work with the government and other AI labs on a voluntary shared standard for frontier-model security and evaluation.
The company said it will create dedicated teams for government priorities, provide compute for government testing and research, and make its safety and red-teaming expertise available for AI evaluation work.
Mythos 5 had already been allowed back for a narrower set of users. The Trump administration recently permitted access for a preapproved list of organizations, including non-US members of those organizations and Anthropic’s own foreign-national employees, according to The Verge. Anthropic said Tuesday it will keep working with the government to widen Mythos 5 access to more domestic and international partners.
Anthropic also said it is working with Amazon, Microsoft, Google, and other companies in its Project Glasswing program on a framework for rating jailbreak severity. The proposed categories include how much capability an attacker gains, how broad that gain is, how easy the technique is to weaponize, and how easily others could reproduce it. Anthropic said it has created a team to monitor jailbreak-reporting channels around the clock and plans to launch a HackerOne program for Fable 5 reports.
This story draws on original reporting from The Verge.
