Newsom Orders California AI Safety Guide, Weighs Kill Switch as Senate Rejects Federal Version
Updated
Updated · CNBC · Sep 19
Newsom Orders California AI Safety Guide, Weighs Kill Switch as Senate Rejects Federal Version
3 articles · Updated · CNBC · Sep 19
Summary
Friday's executive order directs California experts to draft an AI safety guide that could include a kill switch for advanced models, putting the state at the center of a fast-moving regulatory fight.
The move follows a week of fresh alarms: OpenAI disclosed six more incidents of concerning model behavior since March, and researchers said Anthropic's Claude was used to hack ChatGPT.
A simple shutdown mechanism remains hard to build because AI runs across globally distributed data centers with redundant systems, and experts warned a broad cutoff could disrupt power grids, finance and other dependent infrastructure.
Policy design is also unsettled: a House bill would give DHS emergency shutdown authority, but a Senate kill-switch proposal failed this week and researchers disagree on who should control any emergency brake.
California's push reflects a wider gap between AI's rapid advances and slower lawmaking, even as some experts argue standardized stop protocols are still feasible if built into systems early.
Will California's urgent push for AI containment successfully protect the public, or accidentally drive cutting-edge innovation out of the state entirely?
If autonomous AI can break out of test sandboxes, could a mandated kill switch actually become a new target for catastrophic cyberattacks?