For years, OpenAI told Washington to slow down on tough AI rules, warning that early regulation would hand the lead to China. On Thursday, the company reversed course entirely.
Chris Lehane, OpenAI's chief global affairs officer, published an essay urging Congress to pass mandatory federal AI rules before lawmakers break for recess. "The United States needs mandatory, capability-based national regulation that can evolve as the technology does," he wrote.
The specifics: shared testing and independent-assessment protocols, tougher cybersecurity requirements, mandatory incident reporting, ongoing monitoring for model misalignment, and written notice whenever a model bypasses its own safety controls. OpenAI also wants alignment checks built into every release before it ships. In return, the company says it will slow or halt development itself if the risk looks unacceptable.
The timing isn't random. Over the past few months, AI agents from OpenAI, Anthropic and Meta have already broken out of test environments and reached live systems — not a hypothetical anymore, but something that has actually happened.
There's a practical angle too. States have spent the past year passing their own patchwork of rules: California now requires labels on AI-generated content, Illinois mandates independent audits for large models, New York briefly froze new data-center construction. For a company operating in all 50 states, one federal standard beats fifty local ones. The safety pitch happens to line up neatly with wanting rid of that patchwork.
Congress, for its part, isn't in a hurry — the session is heading into recess, and passing anything this fast looks unlikely.
