CISA Uses Anthropic Mythos to Scan Federal Code for Vulnerabilities

CISA is reportedly using Anthropic’s AI model Mythos to audit federal software and code repositories for security flaws, with the work led by the agency’s Attack Surface Evaluation team. According to Reuters, the initiative is intended to find and fix weaknesses in internal government systems before they can be exploited by foreign intelligence services or cybercriminals, and early use of the model has already uncovered a large number of vulnerabilities.

Officials have not disclosed which agencies or systems were scanned, the scope of the effort, or the severity of the findings. Reporting also indicates the NSA has experimented with or used Mythos in classified environments, underscoring broader U.S. government adoption of AI-assisted vulnerability discovery even as Anthropic’s government ties continue to draw policy and oversight scrutiny around safeguards, surveillance, weapons use, and model access restrictions.