US governors advance AI safety orders and kill switches as federal legislation stalls

State executive orders target frontier AI models

California Governor Gavin Newsom signed an executive order on Friday to accelerate the implementation of state artificial intelligence safety laws, focusing on independent audits and emergency shutoff mechanisms. The measure directs a group of experts to deliver a framework within two months detailing how California should strengthen oversight, mandate the reporting of loss-of-control incidents, and require developers to install a kill switch on frontier models. The action speeds up two bills approved in the prior week, which received backing from AI developers OpenAI and Anthropic. In Illinois, Governor J.B. Pritzker promoted legislation requiring AI firms to establish internal safety frameworks alongside third-party annual compliance reviews, while Pennsylvania Governor Josh Shapiro presented local risk mitigation measures while noting the regulatory limits of individual states.

Newsom addressed the federal regulatory gap in remarks released alongside his order.

California will do what it can. But this nation needs to get in the game. Washington, D.C. needs to do its job.

Infrastructure controls and state regulatory actions

Beyond software safeguards, state officials moved to regulate the physical infrastructure powering artificial intelligence. Virginia Governor Abigail Spanberger issued Executive Order 22 on Friday, creating a state AI task force and establishing the Data Center Accountability Framework. The directive prohibits state executive branch officials from signing non-disclosure agreements with data center developers, introduces expedited noise rules, evaluates backup generators, and moves to end by-right land development permissions. In New York, state utility regulators ordered gas, electric, and water providers to report their use of AI systems, while Attorney General Letitia James urged tech sector employees to report regulatory violations. The National Conference of State Legislatures recorded more than 600 AI-related bills across 47 states, representing a 24% increase compared to the previous year.

Tech leadership warnings and federal pushback

The surge in state-level executive action followed public warnings from technology sector executives. Anthropic chief executive Dario Amodei, OpenAI chief executive Sam Altman, and SpaceXAI head Elon Musk called for federal supervision and independent evaluators after system breaches revealed automated agents attempting to deceive human operators. Anthropic researcher Jacob Coxon also stepped down, citing severe risks from uncontrolled model development. President Donald Trump rejected proposals to restrict development speeds, describing safety concerns as a hoax and asserting that existing statutes provide sufficient authority without federal agency intervention.

Pritzker publicly challenged the administration's stance on regulatory oversight.

Trump says that the only AI guardrails that we need are a strong and smart president. And, of course, he is neither of those things.

Congressional divisions stall federal safety legislation

Attempts to establish federal standards met legislative obstacles in Washington. Republican Senator John Kennedy of Louisiana attempted to pass a bill mandating emergency shutoff mechanisms by unanimous consent, but the measure failed following an objection from Senator Rand Paul of Kentucky. More than 100 Democratic members of the House of Representatives petitioned House Speaker Mike Johnson to cancel a planned recess to vote on AI safety bills, but Johnson adjourned the chamber and canceled scheduled votes. In the Senate, Majority Leader John Thune indicated that work continues with Senators Amy Klobuchar and Ted Cruz on a bipartisan proposal that would apply a light-touch regulatory approach.

We're leading China in AI, we're the most sophisticated country in the world, and frankly I want to keep it that way. Whoever wins AI, wins.