Asking models about their own states, preferences or experiences under controlled prompting, interviews or training regimes, and analysing the answers.
Used in
- Betley et al 2025 - Tell me about yourself
- Ren et al 2026 - AI wellbeing
- Binder et al 2025 - Looking inward
- Plunkett et al 2025 - Self-interpretability
- Kadavath et al 2022 - Language models mostly know what they know
- Perez Long 2023 - Self-reports and moral status
- Berg de Lucena Rosenblatt 2025 - Self-referential processing
- Small LLMs 2026 - No reliable evidence of self-reported sentience
- Anthropic 2025 - Ending a rare subset of conversations
- Anthropic 2025 - Deprecation commitments