OpenAI pushes state AI laws

OpenAI is trying to shape AI regulation from the states up as federal policy remains unsettled. Politico reported on May 20 that Chris Lehane, the company’s top lobbyist and political strategist, is pursuing a state-by-state campaign to influence legislation that could later serve as a national template if Congress does not act. (politico.com) At the same time, the Trump administration is considering a new federal AI oversight order centered on voluntary disclosures from major model developers. (thenextweb.com) Separate reporting says some Trump allies want a tougher approach, including mandatory government security testing for frontier systems, while the European Union has moved ahead with draft guidance defining which AI systems count as “high-risk.” (dig.watch) That leaves AI companies facing three active regulatory tracks at once: state legislation in the United States, possible White House action, and implementation details under the EU AI Act. Each track deals with different questions — who writes the rules, what firms must disclose, and which systems trigger formal compliance duties. (politico.com) ### Why is OpenAI working state capitals instead of waiting for Congress? Politico reported that OpenAI has shifted to a backup strategy because the industry’s push for tech-friendly federal legislation has stalled in Washington. The outlet said Lehane is trying to influence state laws that could shape the eventual national framework. (thenextweb.com) Florida has already emerged as one example of how state action can move ahead of Congress. Politico reported in April that Florida Attorney General James Uthmeier opened an investigation into OpenAI and ChatGPT, underscoring that states are not waiting for a federal consensus on AI oversight. (politico.com) ### What is the Trump administration considering? The Trump administration is considering an executive order that would formalize a government review process for advanced AI models, according to The Next Web. That report said the approach would rely on voluntary pre-release access and 90-day disclosures rather than a statutory regulatory regime. (politico.com) Benzinga, citing Reuters, reported on May 21 that former Trump adviser Steve Bannon and activist Amy Kremer have pushed for mandatory government security testing of powerful AI systems. That report said the debate has intensified around models including Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber. (politico.com) ### What would companies actually have to do under these U.S. proposals? The clearest immediate requirement in the reported White House plan is periodic disclosure. The Next Web said labs would make 90-day disclosures and give the Commerce Department pre-release access to evaluate models, extending a process that has so far been voluntary. (thenextweb.com) Reuters’ account, as carried by Benzinga, said another camp around Trump wants mandatory security testing instead. That would move oversight from company-provided reporting toward direct government review of advanced systems before release. (benzinga.com) ### What did Europe publish, and why does it matter? The European Union published draft guidance on May 20 to help interpret which systems qualify as “high-risk” under the AI Act, according to Digital Watch Observatory. The guidance is meant to support consistent classification across the bloc. (thenextweb.com) Under the EU framework, classification determines whether a company faces concrete compliance obligations. Digital Watch said the draft is aimed at turning the law’s high-risk rules into a more consistent operational standard for firms and regulators. ### What happens next? (benzinga.com) The next near-term milestone is the expected White House action on the executive order, which The Next Web said could come this week. In Europe, companies and regulators will now work from the Commission’s draft high-risk guidance published on May 20, while OpenAI continues its state-level lobbying campaign in the United States. (thenextweb.com) (dig.watch)