Image Credit: Shutterstock
Key Takeaways
- The Commerce Department lifted its June 12 export controls on June 30, restoring Fable 5 access starting July 1.
- The original suspension followed a jailbreak that let Fable 5 generate exploit code.
- Mythos 5 access remains limited to roughly 100 approved U.S. companies and agencies under Glasswing.
The U.S. Department of Commerce lifted export controls on Anthropic’s Claude Fable 5 and Claude Mythos 5 models on June 30, clearing the way for Fable 5 to return globally on July 1. The reversal ends an 18-day suspension triggered by a jailbreak that let the model generate exploit code.
Commerce Department Reverses June 12 Order
Anthropic said in a post on X that it received notice the export controls had been lifted and would begin restoring access the following day. Fable 5 is returning across Claude.ai, the Claude Platform, Claude Code and Claude Cowork, with access to Amazon Web Services, Google Cloud and Microsoft Foundry to follow.
Mythos 5, which shares Fable 5’s underlying architecture but carries fewer safety restrictions, is being restored on a narrower basis. The government approved renewed access on June 26 for roughly 100 U.S. companies and federal agencies involved in defending critical infrastructure, through Anthropic’s Glasswing program. Anthropic said it will keep working with the government to widen that access.
A Jailbreak Triggered the June 12 Freeze
The original order landed June 12, three days after Fable 5’s launch. It directed Anthropic to cut off access for any foreign national, including its own non-citizen employees, citing national security authorities. Because the rule took effect immediately and Anthropic had no reliable way to verify every user’s nationality in real time, the company suspended both models for all users rather than risk violating the order.
The immediate trigger was a discovery by Amazon researchers, who found a prompting technique that bypassed Fable 5’s safeguards. The technique, known as a jailbreak, got the model to flag software vulnerabilities and, in at least one case, produce code demonstrating how a flaw could be exploited.
Anthropic’s Safety Commitments
To address the concern, Anthropic said it trained a new classifier designed to detect and block the specific jailbreak technique, which it said now succeeds more than 99% of the time. Requests flagged by the classifier are routed to the less capable Opus 4.8 model instead, with users notified of the switch.
In a letter to Anthropic co-founder Tom Brown, who led the negotiations, Commerce Secretary Howard Lutnick said the company agreed to proactively detect and address model-related safety risks. Anthropic also agreed to coordinate with the government on future model launches and report any malicious use it identifies. Anthropic also opened a bug-bounty program for researchers to report new jailbreaks in Fable 5.
Market Reaction on Hyperliquid
Anthropic is privately held, but traders have taken positions on its valuation through a pre-IPO perpetual, a futures contract without an expiry date, listed on the onchain exchange Hyperliquid.
The contract fell roughly 3.7% when access was suspended in June, a move traders linked to uncertainty over the timing of any future public listing. Pricing on the contract following the July 1 restoration was not immediately available.
Broader Ties to Federal AI Policy
Anthropic said the arrangement includes giving designated government agencies early access to frontier models and their safeguards before public release, along with sharing threat intelligence. The company is also drafting a joint framework with Amazon, Microsoft and Google for scoring the severity of jailbreak techniques.
The episode ties into a June 2 executive order on AI security that created a voluntary path for companies to have frontier models reviewed by the government before release. The arrangement expands the federal government’s role in evaluating frontier AI models before they reach the public.
