Cybersecurity vets protest 'dangerous' US government ban on Anthropic's most powerful models

Key Points:

• A coalition of 76 cybersecurity experts, including industry veterans like Alex Stamos and Katie Moussouris, has published an open letter urging the U.S. government to lift export controls on Anthropic’s Fable and Mythos AI models, arguing these restrictions hinder defenders' ability to find vulnerabilities and improve security.
• The U.S. government recently ordered Anthropic to limit exports of these models citing national security concerns but did not provide detailed reasons; in response, Anthropic suspended access globally.
• Anthropic initially restricted Mythos access to select companies due to its powerful vulnerability detection, and released Fable publicly with strict guardrails preventing cybersecurity-related use, which many experts found overly restrictive.
• The export control order may have been influenced by a non-public Amazon research paper suggesting a method to bypass Fable’s guardrails, but experts like Moussouris argue the paper did not demonstrate a true jailbreak and that the demonstrated techniques reflect normal security testing capabilities.
• The experts call for transparent, scientifically grounded, and democratically developed regulations that balance national security with the need to empower cybersecurity defenders, emphasizing minimal restrictions to protect the American public.