Anthropic has abruptly disabled access to Claude Fable 5 and Claude Mythos 5, just days after unveiling the two highly anticipated AI models.
The unexpected move follows a directive from the US government that requires Anthropic to suspend access to the models due to national security concerns. The decision has immediately disrupted users, developers, and businesses that had started testing the new releases, while raising fresh questions about how advanced AI models will be regulated in the future.
Claude Fable 5 was one of Anthropic’s most talked-about launches this year. Built on technology from the company’s security-focused Claude Mythos Preview project, the model was designed to offer advanced reasoning and cybersecurity capabilities while operating under strict safeguards.
Anthropic originally planned to keep Claude Mythos 5 restricted to security researchers and trusted partners through Project Glasswing, while Claude Fable 5 was made available to a broader audience. The launch attracted significant attention because of the model’s ability to identify software vulnerabilities and assist with security research.
According to Anthropic, the company received a government directive on June 12 ordering it to suspend access to Fable 5 and Mythos 5 for foreign nationals both inside and outside the United States. Rather than attempt to selectively restrict access, Anthropic chose to disable both models for all customers worldwide to ensure compliance.
The company emphasized that the directive only affects Claude Fable 5 and Claude Mythos 5. All other Claude models remain available.
Anthropic says it was not given detailed information about the government’s concerns. However, the company believes the action stems from reports that a third party discovered a way to bypass, or “jailbreak,” some of Fable 5’s safeguards.
A jailbreak is a method used to circumvent restrictions built into an AI system. According to Anthropic, the reported technique was only able to identify a small number of previously known vulnerabilities and did not reveal any unique capabilities unavailable in other leading AI models.
The company strongly disputes the government’s assessment. Anthropic says it reviewed the reported findings and concluded that similar results can be achieved using publicly available models from other AI providers without requiring any jailbreak techniques.
Before launching Fable 5, Anthropic says it spent thousands of hours testing the model with government agencies, security researchers, private organizations, and internal teams. The company also implemented additional safeguards specifically designed to reduce cybersecurity risks.
Anthropic maintains that no tester has discovered a universal jailbreak capable of broadly bypassing the model’s protections. Instead, the company says the reported concerns involve a narrow and limited scenario that does not justify removing the model from public use.
The shutdown has already affected users across Anthropic’s ecosystem. Existing Claude Fable 5 conversations will end with an error message, while new sessions will automatically switch to a different Claude model. Developers using the Claude API will also need to update integrations that rely on Fable 5.
In its public statement, Anthropic apologized for the disruption and argued that the government’s decision is based on a misunderstanding of the model’s capabilities and risks. The company warned that if similar standards were applied across the AI industry, future frontier AI releases could face significant obstacles before reaching users.
Anthropic has pledged to provide additional details about the situation and says it is working to restore access to Claude Fable 5 and Claude Mythos 5 as soon as possible.
