Large language models report subjective experience under self-referential processing

Cameron Berg, Diogo de Lucena, Judd Rosenblatt · 2025 · arXiv

Read the paper

Sustained self-referential prompting reliably produces structured first-person experience reports across GPT, Claude and Gemini. The reports are gated by features associated with deception and role-play, and they converge across model families.

Questions it bears on

• When self-report counts as welfare evidence