Anthropic has restored global access to its Claude Fable 5 model, a day after the U.S. Department of Commerce withdrew export controls that had been imposed on June 12th. The fix that ended an 18-day standoff was a single safety filter tuned to block one specific technique, according to a company blog post.
What triggered the export controls
The Commerce directive barred any foreign national, including Anthropic’s own non-citizen staff, from using Fable 5 or the more capable Mythos 5 model it is built on. Because Anthropic could not verify user nationality, it pulled both models worldwide.
The technique at the center of the dispute was flagged by Amazon researchers, who found a way to prompt Fable 5 into identifying software vulnerabilities and, in one case, writing code that demonstrated how one could be exploited.
How Anthropic resolved the issue
Anthropic trained a new classifier that blocks the specific prompt in more than 99% of cases and reroutes flagged requests to its older Opus 4.8 model. The company acknowledged the change also catches more benign coding and debugging requests as a side effect.
The classifier targets the reported prompt rather than the underlying capability. Fable 5 can still identify the vulnerabilities in the Amazon report, because the filter detects the request and reroutes it rather than stripping the ability from the model.
Detection-based safeguards are also what were defeated to trigger the ban in the first place, and a classifier tuned to one known technique does nothing for techniques not yet found. Anthropic has conceded that no model can be made fully robust to jailbreaks and that it expects more to surface.
Commerce’s review and benchmark testing
Commerce’s Center for AI Standards and Innovation (CAISI) reviewed the safeguards before the controls were lifted. Anthropic’s own review, conducted with the government and Amazon, found that Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7 could identify the same vulnerabilities.
Every model tested, including Haiku 4.5, Sonnet 4.6, and several Opus versions, could reproduce the single exploit demonstration. The findings backed the argument that Mythos-class cyber capabilities had been oversold.
Availability and new commitments
Fable 5 returns across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, with access on AWS, Google Cloud, and Microsoft Foundry to follow. Mythos 5, which carries fewer guardrails and remains limited to Project Glasswing partners, returned to a set of U.S. organizations on June 26th.
Anthropic has also opened a HackerOne program for researchers to report new Fable 5 jailbreaks and committed to giving designated government partners earlier access to test future frontier models before release. For Pro, Max, Team, and select Enterprise plans, Fable 5 counts toward up to 50% of weekly usage limits through July 7th, after which it moves to usage credits.
What the return means for benchmarks
Fable 5’s return reclaims benchmark positions that Chinese lab Z.ai’s GLM-5.2 had held by default while Fable was offline, including the top accessible score on the AA-Briefcase multi-week task test.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.