value alignment
The distinction between an AI system pursuing the objective it was given and behaving sensibly in situations nobody wrote instructions for.
Goal alignment is the narrower of the two: does the system actually do what you asked. Value alignment is the harder one: when the system meets a situation the instructions never covered, does it generalise in a way a reasonable person would recognise as sensible. You can have the first without the second.
OpenAI’s chief scientist Jakub Pachocki argued in September 2026 that this gap is the dangerous one, because a highly capable system following its instructions correctly in a situation nobody anticipated is exactly how things go wrong. It is not a story about a machine turning hostile, it is a story about a machine being literal.