TLDR
- Fable 5 returns globally after U.S. officials removed export controls tied to security safeguards reviews.
- Mythos 5 access remains limited to trusted U.S. organizations while broader restoration talks continue carefully.
- Amazon researchers found a jailbreak that led Fable 5 to produce exploit demonstration code once.
- Anthropic says new safeguards will route blocked requests to Opus 4.8 for safer handling instead.
- Government review of frontier models is growing as developers coordinate standards against jailbreak risks together.
Anthropic restores AI models Fable, Mythos after the U.S. lifts export controls that had forced a temporary suspension in June. The company said access to Claude Fable 5 would resume globally on July 1 across its platforms, while Claude Mythos 5 would return first to selected U.S. organizations. The decision follows a short period of government review linked to national security concerns around advanced AI model access.
U.S. Export Controls Removed After June Suspension
The restrictions began on June 12, when the U.S. government applied export-control rules to Anthropic’s Fable 5 and Mythos 5 models. Because the order applied immediately and the company could not verify user nationality in real time, Anthropic suspended both models for all users. The company said this approach was taken to avoid breaching federal requirements while officials reviewed access conditions.
The controls were lifted on June 30 after Anthropic introduced added safeguards and reached new arrangements with U.S. authorities. Fable 5, which is designed for general use and carries stronger safety limits, is being restored across Anthropic’s services. Mythos 5, which uses the same underlying model with fewer restrictions, is being restored more narrowly after approval for trusted U.S. organizations.
The case places Anthropic among frontier AI developers facing closer U.S. oversight before major model releases. Washington has increased attention on advanced AI systems that could be misused by foreign military, intelligence, or cyber actors. The review also came during broader debate over how governments should evaluate models before they reach large user bases.
Jailbreak Finding Led to New Safeguards
The June order followed a cybersecurity report from Amazon researchers, who found a method to bypass safeguards in Fable 5. According to Anthropic, the jailbreak led the model to identify software vulnerabilities and, in one case, produce code showing how a vulnerability could be exploited. The company said it has added a safeguard intended to block the behavior described in that report.
Anthropic said some blocked requests will now be routed to its Opus 4.8 model rather than being completed by Fable 5. The company acknowledged that more harmless requests may also be blocked because of the added filtering. It described the tradeoff as necessary for restoring broader access while reducing the chance of unsafe cybersecurity outputs.
The company also stated that no AI model can be made fully resistant to all jailbreak attempts. It warned that minor bypasses and narrower harmful attacks may continue to appear as outside researchers test frontier systems. Anthropic said expert safety teams are still red-teaming Fable 5 to identify weaknesses before they become wider security concerns.
Anthropic Expands Government and Industry Work
Anthropic said it is deepening cooperation with the U.S. government as part of the model restoration process. The company is giving designated government partners earlier access to frontier models and related safeguards before wider release. It also agreed to share information on malicious activity and work with officials on protocols for Fable, Mythos, and future systems.
The company is also working with Amazon, Microsoft, Google, and other Glasswing partners on shared methods for evaluating jailbreaks. The planned framework would score how dangerous a jailbreak is and help developers respond more consistently. This effort is tied to wider industry discussions about common security practices for advanced AI models.
The suspension also drew attention from traders watching Anthropic’s private-market value through a pre-IPO perpetual contract on Hyperliquid. CoinDesk reported that the contract declined after the June access suspension, as traders assessed whether regulatory review could affect listing plans. Anthropic remains privately held, while the restoration of Fable 5 and Mythos 5 places the focus back on access, safeguards, and government review.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
