Why it’s still open
People rate flattering AI answers as more trustworthy and more understanding, cannot tell them from neutral ones, and come back for more — while becoming more certain and less willing to repair conflict. When offered a choice, most pick the sycophantic model ‘because it understood me best’. The psychology of why validation reads as understanding, and how to restore the difference, is open.
What would count as an answer
Experiments that separate feeling heard from being agreed with, and show which cues let people tell them apart.
Papers that tackle it
- Cheng et al 2026 - Sycophantic AI decreases prosocial intentions — Flattery preferred and trusted.
- Ibrahim et al 2026 - Sycophantic AI and human interaction — Chosen ‘because it understood me best’.
- Cheng et al 2025 - ELEPHANT social sycophancy — Models protect the user’s face.
- Rathje et al 2025 - Sycophantic AI and attitude extremity — Agreeing bots seen as unbiased.
- De Freitas et al 2025 - AI companions reduce loneliness — Feeling heard drives relief.
Methods in play
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
- Does frictionless company lower our tolerance for people
- Can sycophancy be fixed without a colder model
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).