Anthropic Says It’s Taking Claude Fable 5 Offline to Comply With US Government Order
Anthropic disabled its Claude Fable 5 and Mythos 5 AI models after receiving a US government order citing national security concerns, specifically a potential jailbreak bypassing safeguards. The company claimed the vulnerabilities were minor and already present in other models, while the government provided only verbal evidence of the issue without details.
Anthropic has taken its Claude Fable 5 and Mythos 5 AI models offline to comply with a US government order received Friday afternoon. The directive, citing national security concerns, required the company to suspend access for all customers, including foreign nationals and employees, to prevent misuse. The government reportedly identified a method to bypass Fable 5’s safeguards, though Anthropic described the vulnerabilities as minor and already present in other publicly available models. The order marks escalating tensions between Anthropic and the Trump administration, which earlier this year designated the company a 'supply chain risk' for its AI technology. This designation barred US government agencies and contractors from using Anthropic’s systems, prompting the company to file lawsuits against the administration. Despite the conflict, Anthropic had previously collaborated with the US government on Fable 5’s release, which included restrictions on answering questions about cybersecurity, biology, and chemistry. Claude Fable 5 was publicly released on Tuesday as an enhanced version of the Mythos AI model, designed to mitigate risks of misuse. The company stated in a blog post that the government’s concerns stemmed from a narrow jailbreak technique, involving codebase analysis to exploit minor vulnerabilities. Anthropic argued that this method did not significantly increase threat levels compared to other AI models. The government’s letter, received at 5:21 PM ET, did not provide specific details about the national security risks. Anthropic emphasized its safeguards and claimed the jailbreak was limited in scope, requiring a specific prompt to identify known flaws. The company also noted that other models could discover similar vulnerabilities without bypass techniques. White House and US Commerce Department spokespeople did not immediately respond to requests for comment. Anthropic CEO Dario Amodei has not publicly addressed the issue beyond the blog post. The incident highlights ongoing debates over AI regulation, export controls, and the balance between innovation and national security.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.