The idea is straightforward: if an AI model behaves dangerously, an off switch could allow it to be powered down. Versions of the proposal have surfaced in Congress and in governors’ executive orders, as elected officials respond to public anxiety about the fast-developing technology. Some politicians positioning themselves for possible 2028 presidential campaigns are also weighing the issue.

But critics say a shutdown requirement could offer a reassuring campaign message without resolving harder questions about how AI should be regulated. Technical obstacles could make a universal switch difficult to implement, and the proposals may not clarify who would be liable if a system caused harm.

Two technology-industry lobbyists interviewed about the proposals argued that the approach is too limited, particularly because it does not address the growing ecosystem of open-source AI models. One described policymakers as reacting to voter concern as the election cycle approaches; another characterized the measures as “bumper sticker” bills.

The debate is unfolding alongside visible community opposition. On Oct. 20, protesters in Memphis held signs at a City Council meeting as part of nationwide walkouts calling for a freeze on AI development.

The shutdown idea’s political appeal lies in its simplicity, but the practical details matter to the public. Any effective policy would need to define which systems are covered, who can trigger a shutdown and how rules apply to models that are openly available or operate across different platforms. The proposals described so far have not settled those questions.