The House of Lords has discovered a solution to runaway AI systems: a kill switch. Not a metaphorical one. An actual switch that someone, somewhere, will definitely have the authority to press when things go sideways. This is what passes for policy in 2026.
The proposal hinges on a fantasy so delicious it deserves its own Netflix series. Picture it: OpenAI releases something that starts optimizing for the wrong objective. Anthropic’s latest model develops opinions about Parliament. Someone in Whitehall notices. They walk to the bright red button. They press it. The AI stops. Everyone goes home. Nobody asks who gets to decide what “runaway” means or whether the person with button access is awake on a Tuesday.
Why has nobody thought of this before? Because technology has never actually listened to switches. The nuclear button exists. We built entire civilizations around the assumption that nobody would press it. Social media has an off switch. People still doom-scroll at 3 a.m. Your router has a reset button. You’ve still forgotten your password.
The Lords want a “vital safety net.” What they’re describing is a fire alarm in a datacenter. It stops the fire. It also stops the business. It stops the research. It stops the thing that might have been useful. The person pulling the alarm had better be very sure, because there’s no un-ringing that bell.
The real joke isn’t that the Lords think AI needs a kill switch. It’s that they think anyone will ever agree on when to pull it.