Anthropic asks Newsom to let courts block risky AI models
Anthropic told Gov. Gavin Newsom that California courts should be able to halt dangerous AI models before release.
Objective Facts
Anthropic sent California Gov. Gavin Newsom a letter dated October 15 arguing that the state should be able to go to court to stop the release of AI models that pose a serious risk to the public, according to POLITICO, which obtained the letter, and Reuters, which confirmed it with an Anthropic spokesperson. The letter responds to Executive Order N-9-26, which Newsom signed on September 18 and which directs the Government Operations Agency to study onsite independent auditors, verification of safety plans and a kill switch for frontier models, with recommendations due November 16. Anthropic proposed that the attorney general be able to seek court orders against high-risk models, that independent evaluators be embedded at frontier labs and report risks to the attorney general, and that large developers publish risk reports every six months. It said labs should be able to shut down their models quickly but warned, per POLITICO, that a law built only around shutdowns could give false comfort. OpenAI sent Newsom its own letter about a week earlier that backed shutdown controls and embedded auditors but did not go as far on letting the government block models, POLITICO reported. The governor's office did not respond to Reuters' request for comment.
Left-Leaning Perspective
POLITICO's Chase DiFeliciantonio broke the story and framed it as Anthropic going further than OpenAI in inviting state power over AI models, with detailed coverage of how auditors would be paid and rated. The Just Security Early Edition picked up the POLITICO report and stressed the letter's call for strong oversight and its warning that California's approach could shape other states. The San Francisco Standard, writing after Newsom's September order, questioned whether a kill switch is technically feasible and quoted researchers who doubt it would work against advanced models.
Right-Leaning Perspective
No right-leaning outlet found in this search had reported on the October letter itself. The nearest coverage concerns Newsom's September order. A Fox News live blog that day carried Republican lawmakers' views on AI oversight, with Rep. Ralph Norman saying the technology is too complex for Congress to regulate yet and Rep. Jimmy Patronis arguing for industry self-policing. In the California Globe, editor Katy Grimes questioned Newsom's motives on AI safety and pointed to Anthropic's donations to the California Democratic Party.
Deep Dive
Anthropic's letter is one input into a short review Newsom set in motion on September 18. Executive Order N-9-26 asks the Government Operations Agency, working with the Office of Emergency Services, to report by November 16 on whether California should add onsite auditors, independent checks of safety plans, a frontier model kill switch and loss-of-control incident reporting to its existing law, SB 53, which took effect in January. Newsom vetoed a broader bill with a shutdown requirement, SB 1047, in 2024. Any new mandate would need the Legislature, which POLITICO said is expected to take up AI bills when it returns in about a month. The letter also follows Anthropic's June federal policy paper, which said governments should have legal authority to block or deter deployment of models with catastrophic risk. The coverage gets the core proposal right but leaves gaps. Early summaries, including Just Security's, said Anthropic called for kill switches. That is only partly accurate: POLITICO quoted the company supporting rapid shutdown ability, while Reuters reported that it opposes requiring whole-model kill switches and wants narrower tools. Nothing in the coverage defines catastrophic harm or explains how quickly a court could act, and none addresses open-weight models that cannot be recalled after release, gaps Tech Insider pointed out. Skeptics on the right have focused on Anthropic's political donations and its IPO plans, but no outlet found in this search had published a direct critique of the letter's substance. The next marker is the November 16 deadline for the state's recommendations, followed by whatever bills Sacramento lawmakers draft from them. Federal policy is the larger unknown. ABC7 noted that a December 2025 executive order from President Donald Trump seeks to stop states from enforcing their own AI rules, and the Senate and House proposals Reuters cited have stalled during the pre-election break. Whether the state adopts Anthropic's court-order model or OpenAI's lighter monitoring approach will show how much weight Sacramento gives each lab.