Overview
Severity: MEDIUM | Affected: All US-based AI developers | Category: policy
The United States government has enacted the 'Foundational Model Accountability Act of 2025' (FMAA), a landmark piece of legislation aimed at regulating the development of powerful AI systems. The act mandates that any organization developing a foundational model exceeding a specified computational threshold must undergo rigorous, independent third-party security and safety audits before public deployment. These audits must include comprehensive red-teaming to test for vulnerabilities like jailbreaking, harmful content generation, and potential for misuse in cyberattacks. The National Institute of Standards and Technology (NIST) is tasked with establishing the auditing framework and certifying approved auditing firms. The FMAA represents a significant shift from voluntary commitments to legally binding requirements for AI safety, impacting all major AI labs operating in the US.