U.S. Orders Anthropic to Suspend Foreign Access to Fable 5 and Mythos 5 Over Reported Jailbreak
Summary
The U.S. ordered Anthropic to suspend foreign access to Fable 5 and Mythos 5 over a reported jailbreak that could prompt a model to identify software flaws. Anthropic disputed the finding’s severity and said it is working to restore access.
Key points
- The Commerce Department ordered foreign access to Fable 5 and Mythos 5 suspended under export controls, including for foreign nationals in the United States.
- Anthropic disabled both models to comply; access to its other models was unaffected.
- Anthropic said the government provided verbal evidence of a narrow, non-universal jailbreak involving prompts to inspect a codebase and identify software flaws.
- Anthropic disputed the finding’s severity, saying similar capabilities were available in other public models and used by cybersecurity professionals for defense.
- Anthropic said testers had not found a universal jailbreak and described Fable 5 as using narrow jailbreak resistance and active monitoring.
- The effect of the restrictions on Project Glasswing, which gives selected cybersecurity companies access to Mythos for security work, was unclear.
Article Details
- Event Type
- U.S. government export-control order restricting foreign access to Anthropic AI models after a reported jailbreak raised national security concerns
- Impact
- Anthropic disabled Fable 5 and Mythos 5 to comply with restrictions covering foreign nationals inside and outside the United States; its other AI models were unaffected. The order’s effect on Project Glasswing is unclear. Anthropic disputed the severity of the reported finding, describing it as a narrow, non-universal jailbreak and saying similar capabilities were available in other models.
People
Aaron LevieBox’s chief executive who characterized the directive as a major turning point for AI regulation.Chris McGuireA senior fellow at the Council on Foreign Relations who described the across-the-board restriction as highly questionable.Dario AmodeiAnthropic’s chief executive and the recipient of the Commerce Department directive; Anthropic said it disabled the models to comply.Dean BallA senior fellow at the Foundation for American Innovation who called the government’s move “baffling.”Donald TrumpThe article says he signed an executive order directing federal agencies to bolster cyber defenses and establish a voluntary mechanism for early government access to powerful AI models.Howard LutnickThe Secretary of Commerce who issued the directive ordering Anthropic to suspend foreign access to Fable 5 and Mythos 5.Katie MoussourisChief executive of Luta Security who publicly attributed the issue to “Defense Oriented Prompting,” a security-first approach to AI system instructions.Kirsten DaviesDOD CIO who publicly supported the administration’s action, citing national security and the security of military and infrastructure partners.
Vendors
Anthropicgovernment ordered Anthropic to suspend foreign access to Fable 5 and Mythos 5 after citing a national security concern tied to a reported jailbreak method that could bypass their safety limits.OpenAIthat the capabilities demonstrated were already available in other publicly accessible models, including OpenAI’s GPT-5.5.
Products
Fable 5government ordered Anthropic to suspend foreign access to Fable 5 and Mythos 5 after citing a national security concern tied to a reported jailbreak method that could bypass their safety limits.GPT-5.5capabilities demonstrated were already available in other publicly accessible models, including OpenAI’s GPT-5.5.Mythos 5government ordered Anthropic to suspend foreign access to Fable 5 and Mythos 5 after citing a national security concern tied to a reported jailbreak method that could bypass their safety limits.
Countries
Industries
Artificial intelligenceAnthropic to immediately suspend foreign access to Fable 5 and Mythos 5, its two most advanced artificial intelligence models, citing national security concerns tied to a reported method of bypassing the models’CybersecurityMythos was available to members of Project Glasswing, which allowed selected cybersecurity companies to use the model to identify and address security flaws.