US officials from the Department of Commerce have instructed Anthropic, the maker of the world‑premier Claude AI, to suspend its new model Claude Fable 5 and the accompanying Mythos 5 for all customers.
Anthropic confirmed that the current order required the firm to disable both models immediately to maintain compliance, describing the tool as "too powerful for release" before its public launch.
The decision came just two weeks after the model entered a private preview phase, during which a handful of organisations assessed whether the software was capable of exploiting or hacking digital systems. Officials identified a technique known as jailbreaking that could bypass built‑in safety constraints.
The company noted that the vulnerabilities uncovered appear "relatively simple" and are also present in other publicly available models that do not require a jailbreak. The findings prompted a swift shutdown of the tools.
The halt follows the Trump administration’s public criticism of Anthropic, which added a separate lawsuit against the Pentagon for an order labeling the AI company a supply‑chain risk for US government agencies. A federal judge has ruled that the Pentagon’s directive cannot be enforced, allowing agencies to continue using Anthropic, but the legal dispute remains unresolved.
Experts warn that limiting access to these advanced models may impede the safe development of AI and stifle international collaboration. Gina Neff of Queen Mary University in London said that the incident highlighted uncharted territory for AI governance and could hinder work on cyber‑security safeguards.
The incident underscores the delicate balance between rapid AI innovation and the heightened security demands of the United States, with governments asking for stricter controls over emerging technologies that could be weaponised.
