Anthropic, an artificial intelligence company, announced on Friday that it will temporarily disable its advanced AI models, Claude Fable 5 and Mythos 5, for all users following a directive from the U.S. government. This decision affects foreign nationals both inside and outside the United States and is driven by national security concerns.
Government Directive and Company Response
At 5:21 p.m. ET, Anthropic received an official order to cease access to its models by foreign nationals. The company expressed that there might be a misunderstanding and is actively working to reinstate access promptly. The suspension does not impact other AI models offered by Anthropic.
The government reportedly identified a method to bypass, or ‘jailbreak,’ Claude Fable 5. Anthropic reviewed an example of this technique and noted that it highlighted minor vulnerabilities. These weaknesses, however, are not unique to their models and can be found by other publicly available AI models without requiring a bypass.
Launch and Security Concerns
The suspension comes shortly after the introduction of Claude Fable 5 and Mythos 5. Mythos 5, in particular, is touted for its superior cybersecurity capabilities and remains accessible to a restricted group of cyber defenders and critical infrastructure operators.
Anthropic has implemented robust security measures to prevent the misuse of its AI models in cybersecurity contexts. These include safety classifiers designed to detect and block potentially harmful requests related to cyber attacks and exploitation.
Implications for AI and Cybersecurity
Recently, Anthropic’s Mythos-class model demonstrated the ability to transform newly identified software vulnerabilities into operational exploits within hours, rather than weeks. This capability suggests that leading AI models can quickly weaponize publicly disclosed software flaws.
Anthropic’s Red Team highlighted that a single operator could convert a month’s worth of security patches into working exploits in an afternoon at a low cost, without requiring specialized knowledge. Such capabilities challenge traditional software patching processes.
Despite the U.S. government’s concerns, Anthropic maintains that no universal jailbreak methods exist against its latest models. The company emphasized that its safeguards are more effective than any previously deployed model, based on third-party and internal testing.
Anthropic also argued the need for a transparent and fair statutory process for addressing AI deployment risks, emphasizing that a narrow potential jailbreak should not warrant the withdrawal of a widely deployed commercial model.
Earlier this year, Anthropic faced scrutiny from the U.S. Department of Defense, which labeled the company a ‘supply chain risk’ following its opposition to military applications of its technology. Anthropic has since pursued legal action to challenge this designation.
