US artificial intelligence regulation

How the story developed

On September 11, California enacted the nation's first independent AI auditing framework and chatbot safeguards. Governor Newsom signed a sweeping 13-bill technology and child safety package, including Senate Bill 813 and Assembly Bill 1405, which mandate third-party safety verification for frontier models and establish a formal state registry of credentialed AI auditors. Major AI developers, including OpenAI and Anthropic, supported the California auditing legislation.

The California legislative package also introduced strict regulations for minors. Assembly Bill 1709 prohibits platforms from serving infinite scroll, autoplay, and algorithmic feeds to users under 16, with violations carrying civil penalties of up to $50,000 per affected minor. 'Adam's Law' and Senate Bill 1119 require AI companion chatbots to implement crisis protocols, default time limits, and strict parental controls. Assembly Bill 2 exposes large platforms to statutory damages of up to $1 million per child for negligence, while the laws' age-verification requirements raised concerns among digital privacy advocates.

AI-generated

At the federal level, a wave of departures by safety researchers prompted congressional leaders to accelerate new regulatory frameworks. Over a 72-hour period leading up to September 12, seven senior safety researchers resigned from Anthropic and Google DeepMind. The departing scientists warned that current development trajectories pose an extinction-level threat to humanity.

In response, bipartisan lawmakers began drafting emergency legislation to impose mandatory safety audits on frontier models. The proposed federal guardrails rely on compute thresholds, targeting models trained with more than 10^26 floating-point operations. While Anthropic and OpenAI leadership actively lobbied for development slowdowns to manage recursive self-improvement risks, industry lobbying efforts also focused on preventing hard compute caps. Critics and enterprise leaders argued the regulations serve as a moat to protect incumbents from open-source competition.

AI-generated

By September 13, a bipartisan group of House lawmakers urged Speaker Mike Johnson to cancel an upcoming six-week recess to pass the federal AI safeguards. The lawmakers cited attempts by foreign actors to use models for military purposes and a 10 percent probability of existential risk from advanced AI.

What to watch

• State agencies must develop an emergency 'kill switch' for advanced AI models by November 16.
• The US House faces a narrow legislative window to pass AI regulations before adjourning for the November midterms.
• AI chatbot operators must implement crisis protocols and default time limits by July 2027 under 'Adam's Law.'
• Anyone conducting a covered AI audit in California must be registered and meet strict independence standards by January 1, 2029.
• Departing safety researchers warned that current AI development trajectories pose an extinction-level threat to humanity by 2030.