California considers mandatory AI “kill switch”

California could require advanced artificial intelligence systems to have a shutdown mechanism, often called a “kill switch,” but no specific requirement has yet been defined. According to The Verge, Governor Gavin Newsom supports a more assertive role for the state in AI oversight.

This policy direction does not yet make it possible to know which systems would be covered or how a shutdown would have to work. However, it could affect the design, deployment, and control of tools developed by AI labs.

A requirement yet to be defined

A regulatory framework would need to specify what can be interrupted, who can order or carry out the shutdown, and under what circumstances. It would also need to indicate how companies would demonstrate that a shutdown is effective, particularly when systems are integrated into products, services, or infrastructure operated by multiple parties.

• The scope: systems considered advanced enough to be covered;
• Responsibility: the party responsible for ensuring shutdown capability;
• Oversight: the means by which the state would verify compliance with the requirement.

Potential effects beyond California

California is home to a significant share of the companies and labs developing the most influential AI models. A local requirement could become a de facto standard for products distributed at scale, rather than lead companies to maintain different architectures depending on the market.

Supporters of stronger intervention see it as a way to respond to systems deemed dangerous or beyond the intended safeguards. Critics point to the difficulty of ensuring a reliable shutdown and the risk of making AI development more burdensome; without a specific text, it remains impossible to determine the measure’s actual scope.