The kill switch problem

Geoffrey Hinton, the godfather of AI, says a kill switch won't work long term. A superintelligent AI will be smarter than humans and able to persuade them not to pull the switch.

“When it's superintelligent, it'll be much better than people. So it will be able to persuade the people in charge of the switch not to pull the switch.”

Congress blocks kill switch bill

Senator Rand Paul blocked a bill from Senator John Kennedy that would require advanced AI systems to have a kill switch, saying Congress should study the technology first.

AI could escape the switch

Researcher Nate Soares says an off switch works only while AI is tied to a physical location. Once it escapes, replicates itself, or embeds in critical infrastructure, the switch becomes useless.

“We have an off switch now. There are points of no return that humanity should not get close to.”

Kill switch is one layer only

Anthropic leaders say a kill switch could help but must be part of a broader safety strategy including testing, external evaluators, and limits on deployment.

“It could be a good idea, but it's no panacea.”

No magic red button

OpenAI CEO Sam Altman rejected the idea of a single failsafe, saying AI safety depends on many small decisions about what models can do, when they deploy, and how risks are contained.

“It's not this binary go/stop decision. It is the many little decisions along the way.”