California Sets Two-Month Deadline for New AI Safety Rules
Governor Newsom’s executive order demands a kill switch for frontier models, effectively setting a national standard despite congressional inaction.
Key points
- California’s expert panel has two months to design AI rules, including mandatory audits and a kill switch for frontier models.
- The executive order accelerates existing state laws, effectively making California’s regulations the national standard for the AI industry.
- Critics warn that government mandates may create a false sense of security and reduce the incentive for companies to compete on safety.
California Governor Gavin Newsom has issued an executive order that tasks a panel of experts with designing new oversight mechanisms for artificial intelligence within two months. The directive aims to accelerate the implementation of safety protocols, including the potential creation of a "kill switch" for large-scale models. This move positions the state to act decisively in a regulatory vacuum left by a stalled federal government.
The proposed framework goes beyond current state laws by considering mandatory third-party audits and expanded definitions of safety incidents. Reason Magazine notes that while Newsom frames this as a safety measure, the initiative effectively turns state law into a de facto national standard. This phenomenon, known as the California effect, means that companies operating anywhere in the U.S. must comply with California’s rules to remain viable in the market.
Mandatory audits and kill switches
The expert panel is charged with recommending specific regulatory tools, such as requiring frontier AI companies to host independent auditors in their labs. They will also evaluate the feasibility of implementing emergency shutdown capabilities for powerful models. Additionally, the group will review how to classify loss-of-control events as critical safety incidents, a category that currently lacks clear legal definition.
This initiative accelerates the timeline for existing state bills that establish a framework for certifying independent auditors. However, those bills do not currently require developers to submit to audits. The new executive order seeks to bridge that gap by pushing for stricter enforcement and clearer mandates on company behavior.
State law becomes national standard
Because AI infrastructure and markets are national in scope, California’s regulatory decisions carry weight far beyond its borders. Adam Thierer of the R Street Institute argues that this approach creates a rulebook rather than a guidebook for the rest of the country. Companies are unlikely to ignore these laws because the cost of non-compliance would effectively bar them from accessing the largest tech market in the nation.
Legal experts point out that while constitutional doctrines like the dormant Commerce Clause limit state interference in interstate business, recent Supreme Court rulings have narrowed these protections. Consequently, state-level regulations often stand as the primary check on technology companies when federal legislation is absent or delayed.
Trade-offs in safety regulation
Newsom’s push for regulation follows his 2024 veto of a similar bill, which he then criticized for creating a false sense of security. He warned that over-regulation could stifle innovation and leave the state vulnerable to newer, smaller models that pose different risks. This contradiction highlights the difficulty in balancing safety with competitive advantage.
Critics argue that government-mandated safety standards may lower the bar for innovation rather than raising it. If companies only aim to meet regulatory minimums, the incentive to exceed those standards for genuine safety improvements may diminish. This could result in a baseline of compliance that feels safe to regulators but does not necessarily reflect the highest level of technical security.
