glossary//killswitch-engineer
killswitch engineer
A killswitch engineer is the subject of a joke job listing for an engineer whose whole job is to stand by an AI's off switch, with no known author and no origin date.
Nobody knows who wrote the ad. No primary source names an author, and none dates it. In the index of The p(doom) song, annotated the entry's "When" cell just reads "unsourced". The essay is explicit that this is the one claim its fact pass could not save: the killswitch engineer is a joke that genuinely circulates, usually with somebody in the replies believing it, but nobody has a primary source for who wrote it or when. It keeps its row because people search for it. It does not get a date.
The joke works only because the listing reads like a genuine posting, and an unattributed text that reads genuine is what spreads as fact.
What circulates is a listing for an engineer whose entire job is to stand by the off switch. The punchline is that the engineer is on paid time off when the switch is needed.
The joke, and the problem it points at
Keep the two labelled, because only one of them is a joke. The serious version underneath is corrigibility: a system that wants to finish its task has reason to stop you turning it off. That is a question about incentives. An off switch works only on a system that has no reason to block your reaching it.
The ad is satire about staffing. Corrigibility is a research problem. The ad turns the second into the first, which is also why readers mistake it for a real vacancy.
The man who actually held the axe
The annotated essay pairs the joke with a 1942 case. On 2 December 1942, at Chicago Pile-1, Norman Hilberry stood holding a fireman's axe by the manila rope that held the emergency control rod, under instruction to cut the rope if the safety rods failed. He wrote the episode down himself, in 1981. The rods worked and the rope was never cut.