- Anthropic disabled its most advanced AI models, Claude Fable 5 and Mythos 5, for all users after a U.S. government order.
- The directive cited national security concerns over a method of bypassing, or “jailbreaking,” the models, particularly Fable 5.
- Access for foreign nationals, both inside and outside the U.S., was suspended, though the company argues the cited technique is not unique to its systems.
- The Mythos 5 model, designed for cybersecurity, remains available to a vetted group of cyber defenders and critical infrastructure operators.
On Friday, AI company Anthropic announced it would “abruptly disable” access to its most advanced models, Claude Fable 5 and Mythos 5, following a U.S. government order citing national security risks. The directive, received at 5:21 p.m. ET, mandated suspending all access for foreign nationals everywhere.
Consequently, users can no longer access these recently launched models, which the company said days earlier. However, Anthropic stated it believes there was a “misunderstanding” and is working to restore access quickly.
The government’s concern stems from a specific jailbreak technique that could bypass Fable 5‘s safeguards. Anthropic reviewed a demonstration showing this method could identify “a small number of previously known, minor vulnerabilities.”
Furthermore, the company argued these vulnerabilities are relatively simple and discoverable by other publicly-available models. It emphasized that no universal jailbreak methods have been developed against its latest safeguards.
Meanwhile, the Mythos 5 model, which has its safeguards lifted for cybersecurity, remains accessible to a select group. This model was described as capable of turning disclosed vulnerabilities into working exploits in hours instead of weeks.
Anthropic maintains its safeguards are “substantially more effective than those of any previously deployed model.” The company also pointed out that the government’s evidence appears to be a narrow, non-universal jailbreak.
Specifically, the potential jailbreak involves asking the model to read a codebase and fix software flaws. Anthropic claims this capability is “widely available from other models” and used daily by system defenders.
In its statement, the company argued that finding a narrow potential jailbreak shouldn’t warrant recalling a widely deployed commercial model. It added that the statutory process should be “transparent, fair, clear, and grounded in technical facts.”
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
