Anthropic Disables AI Models Claude Fable 5 and Mythos 5 Following US Government National Security Order

Anthropic, a leading AI developer, has announced the immediate disabling of two of its advanced AI models, Claude Fable 5 and Mythos 5. This unprecedented move comes in direct compliance with an export control directive issued by the US government on Friday afternoon, citing pressing national security concerns. The company clarified that while the order specifically requested the suspension of access for “any foreign national,” it has opted to remove access for all its customers globally to ensure full adherence. This incident marks a significant escalation in the ongoing, often contentious, relationship between Anthropic and the Trump administration. The latest directive is not an isolated event but rather the newest chapter in a series of tensions between the AI firm and the current US government. Earlier this year, the Trump administration's Department of Defense controversially labeled Anthropic a “supply chain risk.” This designation followed Anthropic's attempts to establish clear boundaries regarding the US military's potential use of its sophisticated technology. In response to this classification, which effectively barred government agencies and contractors from utilizing its innovations, Anthropic had previously initiated lawsuits against the Trump administration, highlighting a deep-seated dispute over AI governance and military application. Claude Fable 5, one of the models now offline, was only publicly released this past Tuesday. It represents a specialized version of Anthropic's broader Mythos AI model, meticulously engineered with robust safeguards designed to prevent it from responding to queries related to sensitive areas such as cybersecurity, biology, and chemistry. Prior to its public debut, a limited rollout of the Mythos Preview AI model occurred in April. This initial phase was intended to allow companies and organizations to leverage its potent cybersecurity capabilities to fortify their digital defenses, simultaneously addressing concerns that the technology could be maliciously exploited by bad actors to forge powerful hacking tools. In a blog post published Friday, Anthropic detailed receiving the government's letter at 5:21 pm ET, noting a distinct lack of specific details regarding the national security concern. The company stated its understanding that the government believes it has identified a method to bypass, or “jailbreak,” Fable 5. However, Anthropic presented a counter-argument, asserting that a demonstration of this specific technique revealed it could only identify a small number of previously known, minor vulnerabilities. Furthermore, Anthropic claimed that these vulnerabilities appear relatively simple and that other publicly available AI models are capable of discovering them without requiring a bypass. Anthropic further defended its position, emphasizing the strong safeguards implemented to significantly reduce the likelihood of Claude Fable 5's misuse. The company contended that the “jailbreak” technique reportedly discovered by the US government was narrow in scope and would not render an attacker meaningfully more dangerous than they would be using other existing AI models. According to Anthropic, the government has only provided verbal evidence of a potential narrow, non-universal jailbreak, essentially involving instructing the model to analyze a specific codebase and identify software flaws. Anthropic CEO Dario Amodei has previously voiced support for a fair, structured, and transparent government process for blocking unsafe AI models, a principle the company argues this current action fails to uphold.