According to Anthropic, the government identified what the company described as “a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws.” The company said it reviewed the demonstration and concluded the technique was capable of identifying only a limited number of previously known vulnerabilities.
“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities,” Anthropic said. “These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.”
That assessment has been echoed by dozens of cybersecurity professionals who signed a letter to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross. As of Monday morning, the letter had gathered 76 signatures from security executives, chief information security officers, venture capitalists, and researchers.
The signatories contend that Anthropic’s latest systems are valuable security tools but do not possess exclusive capabilities that justify exceptional restrictions. “Anthropic’s Mythos-class models are quite good at finding flaws and weaponizing exploits,” the letter states. “However, they are not uniquely good at these tasks, and many of the undersigned individuals regularly use other foundation and open-source models for security audits and red-teaming every day.”
The group argues that limiting access to frontier AI models could create challenges for organizations that depend on advanced AI systems to identify software weaknesses and strengthen defenses. In their letter, the experts warned that the government’s action “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”
The debate highlights a growing divide between policymakers and cybersecurity practitioners over how advanced AI systems should be regulated. While the administration has moved to restrict access to powerful models amid concerns about misuse and national security, security professionals argue that many of the capabilities cited by regulators are already available through other commercial and open-source models.
Anthropic has also questioned the broader implications of the government’s approach. “If this standard was applied across the industry,” the company said, “we believe it would essentially halt all new model deployments for all frontier model providers.”
The controversy comes amid increasing competition in artificial intelligence and growing reliance on AI tools for both offensive and defensive cybersecurity work. The cybersecurity experts warned that limiting access to leading U.S. models could have unintended consequences as other countries continue advancing their own AI capabilities.
“China’s advanced AI models are only months behind the best American models,” the letter states. “To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous.”
The clash over Fable and Mythos has become a broader test of how governments balance AI security concerns with the practical needs of organizations defending software and critical systems. With Anthropic’s models offline and industry opposition growing, the outcome could shape future policy decisions surrounding the deployment and regulation of frontier AI technologies.
This analysis is based on reporting from Cybersecurity Dive.
Image courtesy of AFP.
This article was generated with AI assistance and reviewed for accuracy and quality.
