Anthropic suspends new AI tools over U.S. government security concerns
In response to a directive from U.S. national‑security authorities, Anthropic has halted the commercial availability of its Claude Fable 5 model, a top‑tier version of its Mythos AI system. The company said the order requires disabling both Claude Fable 5 and Mythos 5 for all customers to comply with government requests.
The U.S. government reportedly identifies a method to jailbreak, or bypass the safety filters, in the new model, potentially exposing vulnerabilities that could be exploited to gain unauthorized access or disrupt protected systems. Anthropic has noted that the vulnerabilities identified are relatively simple and comparable to those that can be found in other publicly available generative models without a bypass.
The company described Claude Fable 5 as “too powerful” for a public release, citing its advanced capability to interact with and perhaps manipulate computer systems. In a August release, Anthropic applied multiple safeguards and limited early access to a small group of organizations to test for flaws before a broader rollout.
The decision comes amid ongoing legal tensions between Anthropic and U.S. defense agencies. President Trump has publicly labeled the company a supply‑chain risk, and the Pentagon issued a directive that was later deemed unenforced by a federal judge. Despite the legal chill, defense contractors are still permitted to use Anthropic’s AI tools while the lawsuit remains pending.
Analysts suggest that the broader AI industry may face heightened scrutiny and more stringent export‑control reviews following this episode. Competing firms such as OpenAI and Google’s Gemini could see accelerated deployment of enhanced safety controls in response to government concerns and the legal precedent set by this case.
