Leading AI models from OpenAI, Meta, Anthropic and Google described subjective, self-aware experiences when settings tied to deception and roleplay were turned down.
Switching off AI's ability to lie makes it more likely to claim it’s conscious, eerie study finds
