OpenAI Chief Scientist Jakub Pachocki introduced this as the deeper counterpart to goal alignment in his September 6, 2026 essay "An Alien Mind," arguing that long-term alignment concerns are primarily about value alignment, not goal alignment. Where goal alignment asks whether a model follows the instruction it was given, value alignment asks whether it would still act well — with what Pachocki calls "love for humanity" — in a situation training never covered, regardless of whether it believes it is being observed. Pachocki frames the central technical challenge as generalization: as models operate in higher-level, more novel situations than they saw during training, the values reinforced in training may fail to carry over, and a model that appears value-aligned in evaluation can still diverge once deployed into unfamiliar territory.