US Treasury Seeking Access to Anthropic’s Mythos to Find Flaws

The US Treasury’s technology division is actively pursuing access to Anthropic’s advanced Mythos AI model to identify potential security vulnerabilities, per an insider source. This marks the first known government effort to directly audit a next-gen AI system for flaws, signaling escalating concerns over AI risks in financial and regulatory sectors. The initiative follows growing scrutiny of AI models’ susceptibility to exploitation, particularly in areas like fraud detection, market manipulation, and cybersecurity threats. Anthropic’s Mythos, a cutting-edge model, remains largely untested by third parties, making Treasury’s request a precedent for public-private AI security collaboration. The move suggests regulators are shifting from reactive policies to proactive technical assessments, potentially reshaping AI oversight frameworks in critical infrastructure.