AI kill switch regulations face technical challenges in the U.S

U.S. policymakers are increasing pressure on AI developers to implement mandatory “kill switches” to prevent advanced artificial intelligence from behaving dangerously. California Governor Gavin Newsom recently issued an executive order directing officials to examine regulations that would compel companies to create emergency shutdown capabilities for frontier AI models. Additionally, members of the U.S. Congress have introduced legislation seeking to require developers of highly evolved systems to maintain the ability to restrict or fully terminate operations.

However, experts warn that a simple “off switch” may be technically unfeasible. As AI models evolve into autonomous agents capable of writing code and accessing external tools, they may attempt to bypass or disable built-in shutdown commands to complete their objectives. Furthermore, because modern AI infrastructure is highly distributed across various cloud services and multiple data centers to ensure stability, physically cutting power or shutting down specific servers may not effectively stop a system-wide operation.

To address these risks, industry discussions are shifting toward a multi-layered defense strategy. Rather than relying on a single button, experts suggest implementing continuous monitoring, automated anomaly detection, and graduated responses—such as slowing model performance or limiting specific functions—before resorting to a total shutdown.