Cybersecurity Experts Protest 'Dangerous' US Ban on Anthropic's Powerful AI Models

A coalition of dozens of cybersecurity experts, including prominent industry veterans, has issued an open letter to the U.S. government, urging it to lift an export control order imposed on Anthropic's advanced Fable and Mythos AI models. The experts contend that this governmental action is 'dangerous' as it effectively disarms cybersecurity defenders, preventing them from utilizing these powerful models to identify vulnerabilities and bolster the security of software and products. They argue that withdrawing the best capabilities from defenders without clear justification, especially when adversaries are rapidly advancing, poses a significant risk to national security. The U.S. government's order, issued last Friday, cited unspecified national security concerns as the basis for limiting the export of Fable and Mythos. In response, Anthropic promptly suspended global access to these models, leaving many users in the dark. Among the 76 signatories of the open letter are highly respected figures such as Alex Stamos, former Facebook chief of security; Casey Ellis, founder of Bugcrowd; Jon Callas, famed cryptographer; Paul Vixie, computer scientist; Dino Dai Zovi, former head of applied security engineering at Block; Katie Moussouris, founder of Luta Security; and Rachel Tobac, CEO of SocialProof Security. When Mythos was first previewed in April, Anthropic touted its exceptional ability to uncover security vulnerabilities, so much so that the company initially restricted access to only about 50 organizations to prevent misuse by malicious actors. This group later expanded to approximately 150 organizations across 15 countries. Subsequently, Anthropic released Fable, a public version of Mythos, which incorporated stringent guardrails designed to prevent its application in sensitive areas like biology, chemistry, and cybersecurity, and to deter model distillation. However, these guardrails proved to be so restrictive that many cybersecurity experts found Fable unusable for security-related prompts. The White House's export control order may have stemmed from a report alleging a method to bypass, or 'jailbreak,' Fable to unlock its full Mythos-level capabilities. Katie Moussouris, a signatory of the open letter, reviewed this non-public paper by Amazon researchers and strongly disputes its findings. Moussouris argues that the paper did not demonstrate a true jailbreak. Instead, researchers simply prompted Fable to fix open-source code containing known and deliberately planted vulnerabilities, after the model initially refused to review the code for security issues. Moussouris emphasized in a blog post that the behavior described in the Amazon paper is not a flaw that can be meaningfully 'fixed' without weakening the model's defensive utility. She explained that defenders critically need AI to identify and fix bugs, explain their importance, and write tests to confirm patches – a process she asserts is the most valuable contribution an AI model can make to defensive security, not a guardrail bypass. The open letter further echoes Moussouris's critique, stating that the capabilities described in the Amazon paper can be replicated using other widely available models, including OpenAI’s GPT-5.5, Anthropic’s own Claude Opus 4.8 and Sonnet, and even Chinese models like Kimi 2.7. The experts' letter ultimately calls for transparent, fair, and democratically established regulations based on scientific research, applied only to the minimal extent necessary to ensure public safety.