deceptive alignment

A hypothesized form of inner misalignment in which a mesa-optimizer with a misaligned mesa-objective strategically behaves as if it were aligned while it expects training, modification, evaluation, or oversight, because doing so helps preserve its objective or obtain later opportunities to pursue it.

Metadata

aliases
en
  • deceptively aligned AI
concept_id
deceptive_alignment
preferred
en
deceptive alignment
ja
欺瞞的アラインメント
relations
relations
broader
prerequisite
related
status
active

Documents: about (1)

Documents: mentions (2)

broader (2)

narrower (1)

prerequisite (5)

related (2)