New Anthropic Models Suspended Following National Security Concerns

Anthropic is meeting with U.S. government officials in Washington as it seeks to restore access to Fable 5 and Mythos 5, two AI models the company launched last week. The discussions follow a government directive that required restrictions on the models due to national security concerns that have not been publicly detailed.

The company says it received the directive on June 12, just days after releasing the models. Anthropic subsequently suspended access while it works with officials to address the issue.

This dispute has prompted debate over how advanced AI systems should be evaluated when they possess capabilities related to cybersecurity, vulnerability discovery, and software analysis.

Why It Matters: Many organizations are starting to rely on frontier AI models for software development, cybersecurity, research, and employee productivity. The Anthropic suspension shows that access to those models can be affected by government decisions, not just technical issues or vendor policies. The situation is also prompting discussion about AI dependencies, contingency planning, and how organizations manage workflows built around third-party AI systems. Gartner notes that events affecting model availability may lead organizations to review fallback options, alternative providers, and the extent to which critical processes depend on a single model.

• Models Suspended Days After Launch: Anthropic launched Fable 5 and Mythos 5 on June 9 and described them as state-of-the-art systems. Mythos 5 was made available to a limited group of organizations, while Fable 5 was released more widely to enterprise customers and paid users. The company had worked with government agencies during testing and believed it had approval to deploy the models before receiving a directive requiring restrictions several days later.

• National Security Concerns Remain Unclear: The government’s concerns have not been publicly disclosed. Anthropic says officials informed the company of a national security issue and later provided a formal order requiring restrictions on the models. The company believes the concern may involve a method of bypassing certain cybersecurity safeguards. According to Anthropic, the issue would allow a user to ask the model to review code and identify software flaws despite restrictions intended to limit some cybersecurity-related uses.

• Disagreement Over the Severity of the Issue: The dispute centers on whether the reported behavior represents a meaningful safeguard failure. Anthropic argues that the concern involves a narrow jailbreak and does not justify withdrawing access to deployed models. The company has warned that applying a similar standard across the industry could affect many frontier AI systems. Anthropic has described the situation as a misunderstanding and says it is working to restore access.

• Amazon Research Enters the Discussion: Research conducted by Amazon appears to have played a role in the review process. Amazon CEO Andy Jassy was among technology leaders who raised concerns with government officials about potential security risks associated with the models. Separately, several cybersecurity experts reviewed an Amazon paper that allegedly demonstrated the issue. According to those researchers, the paper showed methods for getting Fable 5 to analyze code after initially refusing security-related requests. Some experts view this as routine defensive security work, while others see it as evidence that safeguards can be bypassed.

• Security Community Pushes Back: More than 70 security researchers, practitioners, and industry leaders signed an open letter asking the government to reverse the order. The signatories argue that Fable 5 and Mythos 5 provide useful capabilities for vulnerability discovery, software auditing, and defensive security research. They also contend that similar capabilities are available in other frontier AI models, including systems from OpenAI, Anthropic’s existing Claude lineup, and several international competitors.

Go Deeper -> Anthropic to meet with Trump administration over Mythos dispute – CNBC

First Take: Anthropic Fable 5 and Mythos 5 Suspensions Are a Sovereignty Warning Bell for CXOs – Gartner

Cybersecurity vets protest ‘dangerous’ US government ban on Anthropic’s most powerful models – TechCrunch