Anthropic deploys AI model Mythos inside U.S. intelligence for offensive cyber operations

Anthropic, the AI company behind the powerful Mythos model, has embedded roughly six engineers with the National Security Agency to help the agency adapt the technology for offensive cyber operations, according to sources cited by the Financial Times. The engineers, referred to as “forward-deployed engineers,” are advising on how to tailor the model for specific tasks, including hunting software vulnerabilities and potentially breaching foreign networks. The move marks one of the first known cases of a leading AI lab directly supporting U.S. intelligence in offensive cyber work. A source close to Anthropic told the paper that effective offense is the best way to build effective defense. Anthropic unveiled Mythos in April, describing it as highly skilled at finding software flaws. Access was initially limited to the U.S. government and a select group of organizations. For business leaders, this deployment signals a new chapter in AI’s role in national security—and raises fresh questions about how offensive AI capabilities will be governed.