In brief
Anthropic abruptly suspended its next‑generation AI model, Claude Fable 5, following a government order that flagged potential security breaches. The company cited the discovery of a “jailbreaking” technique that could exploit its safeguards, prompting a halt of all users. Although no specific vulnerabilities were named, analysts note the move highlights growing apprehension about powerful generative models. The decision is amid a lawsuit with the Pentagon over the accessibility of the tools to U.S. government agencies and a broader debate over AI safety and export controls.
Anthropic has suspended the use of its powerful new AI model, Claude Fable 5, after US authorities flagged security concerns days after its public release.
In a statement on its website, Anthropic said it was ordered to prohibit foreign nationals from using Claude Fable 5, which the company described as “too powerful.”
The order requires an immediate disablement of Fable 5 and Mythos 5 for all customers to ensure compliance.
Anthropic points out that U.S. security officials did not pinpoint particular concerns, though they suspect a method to bypass the model’s safeguards—a so‑called jailbreaking technique.
The launch came after a private preview in April aimed at identifying internal vulnerabilities, but officials from finance, technology, and government agencies warned about its potential to exploit and hack computer systems.
Critics argue that the company’s self‑proclaimed “unrelease‑worthy” status may be overstated marketing spin, while academics highlight the need for safe testing and collaboration with governments.
Professor Gina Neff of Queen Mary University, London, noted the model’s capability to exploit defenses 73 % of the time in UK tests, labeling it a step change in cyber‑security risk.
The suspension follows a separate lawsuit against the Pentagon over an order to block U.S. agencies from using Anthropic’s tools, with a judge ruling the directive cannot be enforced while litigation continues.
Federal policy previously described the company’s tools as a “supply chain risk,” a designation rarely given to U.S. firms and indicative of serious security concerns.
