Gavin Newsom advances AI ‘kill switch’ in response to safety fears

Gavin Newsom has signed an executive order outlining plans to require AI labs such as OpenAI and Anthropic to implement a “kill switch”, in the latest sign of growing political concern about the technology.

California’s governor, widely seen as a likely Democratic presidential candidate in 2028, on Friday signed the order to set up a group of experts and officials to work out details of new AI safety measures, including how to legally require AI groups to maintain the ability to shut down their models.

The order signals Newsom is toughening his stance on AI ahead of the midterm elections, which determine control of the US Congress. Fear about the technology and the build-out of data centres have become crucial issues for voters.

“While Washington abdicates its responsibility to protect Americans, California is building on the strongest AI regulatory framework in the nation,” Newsom said.

Calls for a “kill switch” follow recent warnings from leading US AI labs that the pace of AI development ought to be slowed. But experts debate whether technical measures to slow, pause or shut down the models are feasible, and whether sufficiently advanced AI would be able to overcome them.

President Donald Trump has rejected the mounting calls to slow the pace of the technology’s development, arguing the US needed to beat China in the AI race and that concerns about safety stemmed from “negative forces”. He has also tried to block state-level AI regulations across the US.

Other Republicans, including the House Speaker Mike Johnson, urged the public not to panic and echoed Trump’s concerns about losing to China.

Several federal legislators have proposed “kill switch” laws in Congress, but the measures are unlikely to advance before November’s elections reset the legislative agenda.

Recommended

The concern about AI safety follows the resignation of Anthropic’s researcher Jacob Coxon who warned that “the people building AI earnestly believe that it could kill us all by the end of the decade”.

The leaders of some of the main US AI groups — Dario Amodei of Anthropic, Sam Altman of OpenAI and SpaceX’s Elon Musk — last weekend made a rare joint call for the rapid advance of model capabilities to be slowed in the wake of hacking incidents and safety warnings.

Amodei proposed embedding independent evaluators within the labs to monitor their models. Newsom said California — where the leading AI labs are based — would examine making this a legal requirement, as well as having safety plans written by independent third parties.

Newsom also ordered the state government to speed up implementation of two new California AI safety laws that set up a framework for independent evaluators and a state registry for AI auditors.