What's next for the AI safety debate

A week of extraordinary warnings about AI is shifting the fight in Washington from whether to regulate to how far policymakers are willing to go. Why it matters: Washington and industry's failure to put basic guardrails around AI came into plain sight this week, spurring fresh momentum for regulatory action. Lawmakers were quick to re-up their AI bills following a stark warning that went viral from a former Anthropic researcher who said he quit the company over safety concerns. The post thrust the topic into public consciousness with more than 150 million views and fueled debate, but congressional leadership and President Trump have largely been quiet. Those are the players to watch for a true change in the U.S. AI policy approach. What we're watching: The White House and Congress have different cards to play. House Speaker Mike Johnson (R-La.) said at the GOP convention this week that he wants to bring AI companies together to reach "consensus" on safety guardrails, "maybe even at the beginning of the winter," according to remarks his office pointed Axios to. A pause on the development of advanced AI, championed by the likes of Sen. Bernie Sanders (I-Vt.) and likely to come up in a bipartisan briefing next week, would have limited effect without adversary buy-in, according to researchers who argue the U.S. has an obligation to stay ahead of China. President Trump and Chinese President Xi Jinping will meet at the end of this month, with AI safety on the agenda. The White House did not respond to requests for comment. On the Hill, there have been calls for new AI-focused committees, commitments from Democrats to hold hearings should they take back either chamber in the midterms, and plans to mark up existing bills. Between the lines: Without clear support from Senate or House Republican leaders — and by extension, Trump — Congress is poised to keep debating ad nauseam. The Trump administration's new framework for reviewing advanced AI models doesn't require companies to publicly report real-world incidents, as Axios reported this week. The offices of Senate Majority Leader John Thune (R-S.D.) and House Majority Leader Steve Scalise (R-La.) did not respond to requests for comment. The offices of Senate Minority Leader Chuck Schumer (D-N.Y.) and House Minority Leader Hakeem Jeffries (D-N.Y.) also did not respond. Zoom out: The debate in the U.S. is unfolding while other governments around the world move ahead with AI rules. The EU has already put some basic guardrails in place. The EU AI Act requires providers of the most powerful AI models to report serious incidents to regulators, and new transparency and disclosure requirements are now in effect for chatbots and AI-generated content. China has also imposed rules on AI chatbots, including restrictions on virtual companions for minors and requirements to label AI-generated content. Beijing has also called for greater international cooperation on AI safety. Friction point: One factor is the tech industry's influence on lawmakers. OpenAI's top lobbyist Chris Lehane this week said this moment requires a "new" chapter on AI policy, but the company's positions are so far largely the same. His first point talks about working with Congress on mandatory rules — in other words, a federal law. That's something the company has long supported, as one national standard would be easier to tackle than playing whack-a-mole at the state level, where AI bills can actually pass. Lehane writes of eventually backing legislation in Congress and getting something across the finish line as soon as this year, but doesn't specify which one. The FRONTIER Act, from Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.), would impose tiered requirements based on the size of AI companies in an attempt to address concerns that regulation could favor the biggest players. A separate proposal in the works from Thune and Sens. Amy Klobuchar (D-Minn.) and Ted Cruz (R-Texas) is expected to leave it up to companies to test their own models for catastrophic risks and report the results to the government. The bottom line: While AI is moving faster than Congress, it's clear that basic measures like incident reporting are overdue.

Read it at Axios ↗

← Front page