It sits between capability research and product safety: papers on specification gaming, scalable oversight, and deceptive alignment are alignment research even when they do not ship a filter. Labs fund it because failures at deployment are expensive, and because future systems may be harder to evaluate by inspection.