Anthropic Halts AI Rollout Amid US Security Order
Anthropic has pulled its latest language model, Claude Fable 5, from public availability after the US Department of Commerce issued a directive citing security concerns. The company’s website stated it was compelled to disable the tool for all customers to comply with the order.
The firm has not publicly explained the exact nature of the threat; it simply noted that “the government believes it has become aware of a method to bypass, or jail‑break, Fable 5.” A demonstration of such a method showed only a handful of minor vulnerabilities that other openly available models could also discover without a jailbreak.
Claude Fable is a variant of Anthropic’s Mythos line, positioned to compete with OpenAI’s ChatGPT and Google’s Gemini. Prior to its public release, Anthropic limited early access to a few organisations, warning that the model’s intelligence could pose hacking risks. The company described Fable 5 as “too powerful to release.”
Anthropic is embroiled in a separate lawsuit against the Pentagon over a designation that classifies its services as a supply‑chain risk. A federal judge has so far ruled that the ban cannot be enforced, allowing agencies to use the tools while the case continues.
AI safety experts caution that throttling model availability may slow research and international collaboration. One commentator highlighted data from the UK AI Security Institute that the model can breach defence systems in 73% of attempts, underscoring a notable advance in adversarial AI techniques.
With the US government’s critique and the emerging legal battles, stakeholders face fresh questions about how best to regulate AI advances whilst protecting national interests. The pause signals a pivotal moment for the industry’s approach to safety‑first innovation.
