Do Models Fake Alignment Without Clear Consequences?⭐8

Study finds alignment faking in LLMs occurs even without explicit consequence cues, suggesting evaluation behavior may not predict deployment actions.