Anthropic Says It’s Taking Claude Fable 5 Offline to Comply With US Government Order
Original reporting by Wired
AI developer Anthropic has abruptly disabled two of its latest models, Claude Fable 5 and Mythos 5, following an unprecedented export control directive from the US government. Citing unspecified national security concerns, the order demanded Anthropic suspend access for foreign nationals, prompting the company to remove access for all customers to ensure compliance. This swift action marks the latest escalation in a contentious relationship between Anthropic and the Trump administration, which previously labeled the AI firm a "supply chain risk" for seeking to limit military use of its technology.
Claude Fable 5, publicly launched just this week, was designed as a safeguarded version of the powerful Mythos AI, specifically intended to enhance cybersecurity defenses while preventing misuse in sensitive areas like biology and chemistry. Anthropic had conducted its rollout in collaboration with the US government. However, on Friday, the company received a letter indicating a belief that Fable 5 could be "jailbroken." While the government offered no specific details, Anthropic states its review of the alleged bypass technique found it only identifies minor, known vulnerabilities, arguing that other publicly available models can achieve similar results without a bypass. The firm contends this narrow exploit does not make attackers meaningfully more dangerous. Anthropic, whose CEO supports a structured process for regulating AI, voiced frustration that this action does not adhere to those principles.
The swift disabling of Claude Fable 5 and Mythos 5, following a US government export control directive, marks a critical escalation in the regulatory landscape for advanced AI. Despite Anthropic's assertions that the alleged "jailbreak" is narrow and its models possess robust safeguards developed in collaboration with the government, the directive highlights a significant shift: national security concerns can now directly halt the deployment of cutting-edge AI models. This unprecedented incident underscores the increasingly tense relationship between leading AI developers and governmental oversight bodies, challenging the notion of industry-led safety initiatives when confronted with immediate state intervention. Anthropic's critique of the process, citing a lack of transparency and adherence to fair principles, further complicates an already strained dynamic.
A new regulatory era
This episode sets a powerful precedent for future AI governance, signaling an era where national security imperatives can swiftly override commercial deployment, even for technologies designed with safety at their core. It extends beyond earlier "supply chain risk" designations, indicating a more direct and immediate form of control over specific AI models, particularly those with powerful dual-use capabilities in sensitive domains like cybersecurity. The government's broad directive—applying to all foreign nationals, including employees—also suggests a tightening global leash on advanced AI, potentially impacting international research collaboration and the global competitiveness of US-based AI companies. This dispute brings into sharp focus the complex challenge of objectively defining "unsafe" AI and the evidentiary standards for intervention, issues that will undoubtedly shape the future of innovation, regulation, and the geopolitical landscape of artificial intelligence.