Why a 'kill switch' isn't the right approach for AI regulation

In response to recent high-profile hacking incidents involving advanced AI systems, some politicians and business leaders have proposed the use of an AI kill switch—a mechanism that could shut down AI systems in emergencies. However, experts argue that this metaphor is flawed. Dylan Baker, a research engineer at the Distributed AI Research Institute, explains that modern AI models rely on complex, distributed technologies that cannot be controlled by a single kill switch. The concept also overlooks the human decisions involved in AI oversight. The kill switch metaphor, like the black box metaphor, simplifies AI’s complexity, framing it as an incomprehensible entity. While several 2026 regulatory efforts have focused on kill switches, including the AI Kill Switch Act and California’s executive order, no national regulation has included this feature. President Donald Trump has opposed the idea, calling it unnecessary regulation.