Anthropic has restored access to ‘Claude Fable 5‘ worldwide after what it described as productive discussions with the US government over cyber security safeguards and the model is returning with additional protections and some temporary feature limits. The company said it is redeploying the model with new classifiers designed to detect and block more cybersecurity-related tasks, while routine coding and debugging requests may temporarily fall back to Opus 4.8.
The company also added that access to Claude Fable 5 and Claude Mythos 5 had been suspended after US export controls were applied on June 12, requiring Anthropic to restrict access to foreign nationals, including those inside and outside the United States. “Because the order took effect immediately and we had no reliable way to verify nationality in real-time, we suspended access to both models for all users,” the company said.
Anthropic said the export controls on Fable 5 and Mythos 5 were lifted on June 30 (Tuesday). Fable 5 is now available globally on the Claude Platform, Claude.ai, Claude Code and Claude Cowork, while users on Pro, Max, Team and select Enterprise plans can access it for up to 50 per cent of weekly usage limits through July 7 before it shifts to usage credits. The company also said it will re-enable access on Amazon Web Services (AWS), Google Cloud and Microsoft Foundry as quickly as possible.
Fable 5 back with extra safety features
The renewed rollout follows concerns raised after Amazon researchers reportedly found a way to bypass Fable 5’s safeguards. Anthropic said its own testing confirmed that several less capable models, including Claude Opus 4.8, GPT-5.5 and Kimi K2.7, could identify the same vulnerabilities. It added that the reported technique did not reveal any unique Mythos-level cyber capabilities. To address the issue, Anthropic said it built a stronger classifier to target the specific behavior described in the report.
“Users will be notified if a request to Fable 5 is blocked, and the request will instead be sent to Opus 4.8,” the company said. It added that the new classifier blocks the reported technique in more than 99% of cases, though it may also flag some legitimate coding and debugging tasks more often than before.
Why Anthropic is tightening controls?
Anthropic said Mythos 5 remains the more capable model for finding and exploiting software vulnerabilities, making it especially attractive to malicious actors. Fable 5, by contrast, was launched with what the company called its strongest safeguards yet, including a defense in depth approach that combines model training, post-hoc monitoring and smaller automated classifiers.
The company said it intentionally set a wide safety margin so that the system would err on the side of blocking possibly risky requests, even if some were benign. That approach, Anthropic said, can create false positives, but it also makes jailbreaks harder to achieve and limits the chance that a narrow bypass leads to dangerous misuse. The company said it expects more refinements over the coming weeks as it works to distinguish harmful cyber use from legitimate defensive activity.
Industry framework for jailbreaks
Anthropic also added that the episode highlights the need for a shared industry framework to assess AI jailbreak severity in a more objective way. Working with Amazon, Microsoft, Google and other Glasswing partners, it wants to develop a common standard that would help developers, governments and security teams decide how urgently to respond when a bypass is discovered.
The company’s draft framework would assess jailbreaks across four dimensions: capability gain, breadth of capability gain, ease of weaponisation and discoverability. In practical terms, that means judging how much extra power the jailbreak gives an attacker, how many tasks it works on, how difficult it is to turn into an attack and how easy it is for others to find and reuse. Anthropic said the framework would help calibrate responses to future discoveries, including fast mitigations for the most severe cases, and it is launching a HackerOne program so researchers can submit cyber jailbreaks for review.
Anthropic said it has also deepened coordination with the US government on frontier Artificial Intelliegnec security, including work with the Office of the National Cyber Director, the Office of Science and Technology Policy, the Department of Commerce, the Department of the Treasury and national security agencies. It plans to give designated government partners expanded pre-release access to advanced models and their safeguards, share information quickly when major misuse patterns emerge and dedicate staff, compute and research support to joint evaluation work.
The company said it also wants to help establish a common voluntary security standard for frontier model providers. “These rules should be codified in strong regulation and applied equally across frontier model developers,” Anthropic said, arguing that government involvement in AI releases should follow a durable and transparent process. For users, the immediate takeaway is that Fable 5 is back, better guarded and now positioned as both a more available and more tightly controlled frontier model.