OpenAI Publishes Blueprint for U.S. Frontier AI Safety Framework

OpenAI announced on its website a blueprint detailing how the United States can establish a durable framework for governing frontier AI systems. The plan emphasizes building a national structure that aligns with emerging state laws, strengthening the Center for AI Standards and Innovation (CAISI), and mobilizing a broad government strategy to address national security and public safety risks.

The document outlines a three-part strategy for democratic governance of frontier AI. It proposes creating a national frontier safety framework based on existing state legislation such as California’s SB 53, New York’s RAISE Act, and Illinois’s SB 315. The framework would mandate severe risk evaluations, transparency reports, independent audits, and whistleblower protections for companies developing advanced AI models.

OpenAI recommends expanding CAISI’s authority to become the primary U.S. institution for evaluating and certifying frontier AI systems. The blueprint suggests that CAISI should have resources for technical assessments, access to classified computing environments, and the ability to coordinate with national security and scientific agencies. It also calls for mandatory evaluations of the most capable AI models before release.

The plan further proposes a government-wide resilience strategy to ensure defensive capabilities keep pace with AI advancements. This includes strengthening cybersecurity, biodefense, and critical infrastructure protections, maintaining U.S. leadership in advanced compute resources, and prohibiting federal use of unevaluated frontier AI systems. The blueprint positions democratic governance as essential to managing the risks and benefits of increasingly capable AI technologies.