Congress Has Days to Act on AI Safety. It May Head Home Instead.

Key points

• A researcher who resigned from Anthropic warned in 2026 that AI could "kill us all by the end of the decade," prompting public calls for restraint from the chiefs of Anthropic, OpenAI and xAI.
• House Speaker Mike Johnson signalled Congress would not hold an emergency session to pass AI safeguards before leaving for midterm campaigning.
• Rep. Sam Liccardo led a group of Democrats writing to Johnson demanding the House stay in session until it passes meaningful AI safety rules.
• Senate Majority Leader John Thune is working on a bipartisan AI bill with Senators Klobuchar and Cruz, but it has not been formally introduced.
• An industry source told CNBC that legislation with "real momentum" could emerge within the week.

The people who build the most powerful AI systems in the world are asking the government to stop them. That sentence would have sounded strange two years ago. Right now it is the political reality Washington cannot agree on how to handle.

The trigger was a resignation. A researcher left Anthropic, a San Francisco AI company that makes the Claude family of AI assistants, and went public with a stark warning: the technology could "kill us all by the end of the decade." Within days, the chief executives of Anthropic, OpenAI and xAI (Elon Musk's AI company) each made public pledges to slow development. For an industry that has spent years insisting self-regulation is enough, it was a striking reversal. As we reported on 13 September, the White House and Congress had only recently told those same executives to keep building faster.

What is Congress actually doing about it?

Not much yet, and the clock is running out. Johnson plans to send the House home after this week until November's midterm elections pass. A group of Democrats led by Rep. Sam Liccardo of California wrote to Johnson demanding the House stay in session. "To our children who will have read a post-apocalyptic history, 'Why Congress Slept,' our inaction will be inexplicable, and unforgivable," the letter read.

Johnson said no to an emergency session. His argument: move too fast and the United States hands the advantage to China. "If Congress just races in and does some sort of emergency session to try to regulate AI, we will lose the race to China," he said Sunday on CNN.

Sen. Ruben Gallego of Arizona pushed back. "AI does not care once it becomes fully dangerous whether we're Chinese or American," he said. "That's the danger of this."

President Trump, speaking briefly in Ireland on Sunday, echoed the China concern and cast doubt on the severity of the warnings. "We can put guardrails, and we can do this and that, but I think you have a lot of negative forces that are bringing it up and they're bringing up things that won't happen," he said, according to a pool report.

What proposals are actually on the table?

Several bills already exist in the House. The FRONTIER Act, a bipartisan bill led by Reps. Jay Obernolte and Lori Trahan, would require safety evaluation and transparency for frontier AI models (the largest, most capable systems). Other proposals include mandatory "kill switch" requirements and a temporary antitrust waiver so companies could coordinate on safety without fear of prosecution. We've covered the FRONTIER Act twice since 14 September, each time finding the bill stalled short of a floor vote.

In the Senate, Majority Leader John Thune is working with Senators Amy Klobuchar and Ted Cruz on a bipartisan measure, but it has not been formally introduced. House Democratic Leader Hakeem Jeffries said Democrats would meet Tuesday to discuss next steps.

An industry source familiar with the discussions told CNBC that legislation with real momentum could emerge within the week. Lawmakers facing November ballots, the source noted, would rather point to an accomplishment than a shrug.

Here's my honest read after following this beat: the warnings from the industry's own leaders have opened a narrow window that Washington may simply let close. Johnson isn't wrong that rushed, bad law could do damage. Gallego isn't wrong that waiting for a committee framework that arrives three years from now may be too late. Neither side has offered a plan that clears both of those bars, and the midterm recess is a week away.

Common questions

What would an AI "kill switch" actually do?

A kill switch requirement would legally oblige AI companies to build a mechanism that can shut down or halt a powerful AI system if it begins behaving dangerously. Think of it as a circuit breaker for software that no one fully controls.

Does any of this affect AI tools people use today, like ChatGPT?

The proposals in Congress focus on frontier models, the largest systems under active development, not the consumer chatbots most people use now. Everyday tools would be indirectly affected only if the companies that make them were required to slow the development of future, more powerful versions.