Anthropic has restored global access to Claude Fable 5, one day after the U.S. Department of Commerce withdrew the export controls it imposed on the model on June 12. The move ends an 18-day standoff in which Anthropic had pulled the model and its underlying Mythos 5 system because it could not verify the nationality of its users. The fix that satisfied regulators was a single safety filter tuned to block a prompting technique flagged by Amazon researchers, which Commerce’s Center for AI Standards and Innovation (CAISI) reviewed before lifting the controls.<\/p>\n
The June 12 directive barred any foreign national, including Anthropic’s own non-citizen staff, from using Fable 5 or the larger Mythos 5 model. Because Anthropic had no practical way to verify the nationality of users across its products, it pulled both models worldwide.<\/p>\n
The technique identified by Amazon researchers could prompt Fable 5 into recognizing software vulnerabilities and, in one instance, generating code demonstrating how a vulnerability could be exploited. Anthropic trained a new classifier that blocks that specific technique in more than 99 percent of cases and reroutes flagged requests to the older Opus 4.8.<\/p>\n
Two caveats stood out in the company’s account:<\/p>\n
The filter also targets the request, not the underlying capability. Fable 5 can still identify the vulnerabilities described in the Amazon report, but the request is intercepted and rerouted rather than the ability being stripped from the model.<\/p>\n
Anthropic’s review, conducted with the government and Amazon, tested several frontier models on the same task. The review found that Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7 could identify the same vulnerabilities, and that every model tested, including Haiku 4.5, Sonnet 4.6, and multiple Opus versions, could reproduce the single exploit demonstration. The findings supported the argument that the cyber capabilities of the Mythos class had been overstated.<\/p>\n
Fable 5 returned to Claude.ai, the Claude Platform, Claude Code, and Claude Cowork on the day of the announcement, with access through 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 26.<\/p>\n
During the outage, Chinese lab Z.ai’s GLM-5.2 had held several top positions by default, including the highest accessible score on the AA-Briefcase multi-week task test. Fable 5’s restoration allows direct comparison on those benchmarks again.<\/p>\n
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 usage counts toward up to 50 percent of weekly limits through July 7, after which it moves to usage credits.<\/p>\n
The U.S. Department of Commerce imposed export controls on Fable 5 on June 12 that barred any foreign national, including Anthropic’s own non-citizen staff, from using the model. Because Anthropic could not verify the nationality of users across its products, it pulled Fable 5 and the larger Mythos 5 model worldwide.<\/p>\n
Anthropic trained a new classifier that blocks the specific prompting technique flagged by Amazon researchers in more than 99 percent of cases and reroutes flagged requests to the older Opus 4.8. Commerce’s Center for AI Standards and Innovation (CAISI) reviewed the safeguard before the controls were withdrawn.<\/p>\n
Fable 5 returned to Claude.ai, the Claude Platform, Claude Code, and Claude Cowork on the day of the announcement, with access through AWS, Google Cloud, and Microsoft Foundry to follow. Mythos 5, which carries fewer guardrails, is limited to Project Glasswing partners and returned to a set of U.S. organizations on June 26.<\/p>\n