Newsom Orders AI "Kill Switch" and Independent Auditors in California Executive Order
Governor Gavin Newsom's executive order gives an expert panel two months to recommend rules for shutting down dangerous AI models and placing auditors inside labs, citing a federal vacuum on incident reporting.
California Governor Gavin Newsom signed an executive order on Monday that directs state officials to develop a framework for an emergency "kill switch" on AI models and to push for independent auditors to work inside AI labs. The order sets a two-month deadline for an expert panel to deliver recommendations. Newsom framed the move as a response to the absence of federal requirements for AI companies to report dangerous incidents.
The executive order does not itself impose binding rules on AI developers. Instead, it tasks the panel with proposing how the state could require labs to disclose safety failures and how a shutdown mechanism might work in practice. That approach gives Newsom a way to shape AI oversight while avoiding a direct legislative fight, at least for now. California is home to many of the largest AI companies, including OpenAI, Anthropic, Google and Meta, which gives the state outsized leverage over industry norms.
Newsom's emphasis on a reporting gap is notable. No federal statute currently compels AI developers to notify regulators when a model behaves dangerously, whether through cyberattacks, biohazard assistance or other harmful outputs. The executive order attempts to fill that vacuum at the state level, though its reach is limited to California-based operations and companies doing business there. The panel's recommendations could become the basis for future legislation or emergency regulations.
The "kill switch" language is likely to draw the most attention. In practice, shutting down a deployed AI model is technically and legally complicated, especially when models are distributed across cloud providers and third-party applications. Auditors embedded inside labs would also raise questions about trade secrets, liability and how much access outside experts should get to model weights and training data. The order leaves those details to the panel.
Industry reaction so far has been muted, with major labs declining to comment or saying they are reviewing the order. The two-month timeline means recommendations could land before the end of the year, potentially setting up a clash with federal agencies that have taken a lighter-touch approach to AI safety. What remains unclear is whether Newsom will turn the panel's advice into enforceable rules, and whether other states will follow California's lead or wait for Congress to act.