US Orders Anthropic to Suspend Fable 5 AI Model Over National Security Concerns
Anthropic announced on Friday that it is shutting down access to its advanced AI models Claude Fable 5 and Mythos 5 after receiving a government order at 5:21 p.m. ET citing national security concerns. The order requires blocking foreign nationals from using these models both inside and outside the United States. The company characterizes this as a misunderstanding and is working to restore access, while emphasizing that its other AI models remain unaffected by the shutdown.
The government's action stems from concerns about bypassing Fable 5's security features. According to Anthropic, officials demonstrated a method that exploits minor known weaknesses in the model, though the company contends that other publicly available AI models can identify the same vulnerabilities without requiring any bypass techniques. The timing is particularly notable since these models were just launched days earlier, with Mythos 5 being positioned as having the strongest cybersecurity capabilities globally and remaining available to a limited group of vetted professionals.
The core concern revolves around how quickly advanced AI models can weaponize software vulnerabilities. Anthropic revealed that its Mythos-class models can develop working exploits from newly disclosed security flaws in hours or minutes rather than weeks. The company's Red Team demonstrated that a single individual could create exploits for an entire month's worth of security patches in one afternoon for just a few thousand dollars without specialized skills. This capability fundamentally challenges the effectiveness of traditional monthly patching cycles and gradual security update rollouts.
Anthropic defended its safety measures, stating that it has implemented strong safeguards including classifiers designed to detect and block misuse attempts related to cybersecurity attacks, exploit development, and security bypasses. The company argues that no universal bypass methods exist for its latest models and that their safeguards are more effective than any previously released model. However, they acknowledge that perfect resistance to jailbreaking is impossible for any AI provider. Anthropic maintains that the government has only provided verbal evidence of a specific, limited bypass involving code review requests, and that similar capabilities are already widely available in competing models like OpenAI's GPT-5.5. The company is calling for a transparent and fact-based legal process rather than immediate recalls based on limited potential vulnerabilities.
Stay secure — stay Wavasec. 🔐
