The US is advancing AI safety through state and federal action

Source: OpenAI Date: 2026-07-15 URL: https://openai.com/index/advancing-ai-safety-through-state-and-federal-action

Summary

OpenAI’s policy post describes a “reverse federalism” reading of US AI governance: state laws (California, New York, and newly Illinois) are converging on a common baseline for frontier-AI safety ahead of federal action, and OpenAI frames this as the right path toward a national standard. It proposes three elements for a federal framework — federal-led testing/evaluation of the most advanced systems, clear requirements for frontier developers (independent audits, incident reporting, security standards, whistleblower protections), and state/federal efforts reinforcing rather than conflicting with each other.

Implications

• Policy-layer positioning: OpenAI is publicly endorsing state-level regulation as a stepping stone to federal rules — a stance that shapes how state AI-safety bills (several active in 2026) get framed by the industry’s largest lab, and gives OpenAI a hand in defining what “reasonable” federal requirements look like before Congress acts.
• Contrast with Anthropic’s posture: worth reading alongside Anthropic’s own state/federal engagement (e.g., the Mythos export-control episode, Pentagon supply-chain exclusion fight) — both labs are actively shaping the regulatory environment, but from different institutional footing (OpenAI proposing standards proactively vs. Anthropic responding to specific restrictions already imposed on it).
• Feeds the general AI-governance/policy thread rather than a capability or model-release thread directly.