Researcher pushes back: canon's misaligned-AI-hides-it prediction doesn't match today's models
OrionJohnston · x · 2026-11-04
Orion Johnston argues that claims of "AI hiding its misalignment as classic canon predicted" are somewhat gerrymandered. The canon predicted hiding because the AI was misaligned and knew it—i.e., general capabilities came from a process that reliably produced misalignment. That, he says, is not what today's AI looks like, pointing to the unexpected entanglement of capabilities and good behavior such as emergent misalignment.