Roko's basilisk

A user called Roko posted the argument to LessWrong in July 2010, and it runs like this: a future benevolent AI might punish simulations of everyone who knew it could exist and did not help build it, so merely learning about the idea puts you at risk. None of that is a prediction. It carries no mechanism and no date. It threatens only a reader who has already signed up to a particular way of reasoning about future agents. Pascal's wager works the same way, using a small probability of an unbounded payoff to make one bet look compulsory.

What made it famous was the reaction, not the argument. Yudkowsky deleted the post and banned discussion of the topic on LessWrong for five years, and the ban did what bans do.

The objection on the record

The index in The p(doom) song, annotated gives the basilisk a row dated July 2010 and carries osmarks' classification of it: a rationalist mugging, meaning a bad argument that coerces you with its own stakes. That is the one objection on the record here, and it belongs to osmarks rather than to the essay or to the field at large.

Neither source gives a day in July 2010, and no real name for the poster is on the record. Anyone writing the term down is stuck with the vague version of both, which is the honest version anyway.

The term also travels on its reputation. It arrives in general coverage as shorthand for what a particular community is supposed to be frightened of, which is a different thing from engaging the argument, and worth knowing before you use the word.