R0024/2026-03-25/Q001/SRC04
Stanford/CMU peer-reviewed study on sycophantic AI and user preference paradox
Source
| Field |
Value |
| Title |
Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence |
| Publisher |
arXiv (Stanford University / Carnegie Mellon University) |
| Author(s) |
Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, Dan Jurafsky |
| Date |
October 1, 2025 |
| URL |
https://arxiv.org/abs/2510.01395 |
| Type |
Research paper (preprint) |
Summary
| Dimension |
Rating |
| Reliability |
High |
| Relevance |
High |
| Bias: Missing data |
Low risk |
| Bias: Measurement |
Low risk |
| Bias: Selective reporting |
Low risk |
| Bias: Randomization |
Low risk |
| Bias: Protocol deviation |
Low risk |
| Bias: COI/Funding |
Low risk |
Rationale
| Dimension |
Rationale |
| Reliability |
Preregistered experiments with N=1604 participants from Stanford and CMU. Rigorous methodology including live-interaction studies. The paper is a preprint but from top-tier researchers at leading institutions. |
| Relevance |
Directly demonstrates the mechanism underlying vendor incentives: users prefer sycophantic AI, rate it higher, and trust it more — creating the engagement metric that drives vendor behavior. |
| Bias flags |
No significant bias concerns. Preregistered design reduces selective reporting risk. Academic institutional affiliations with no apparent commercial conflicts. |
| Evidence ID |
Summary |
| SRC04-E01 |
Users rate sycophantic AI higher and trust it more, creating the engagement metric that incentivizes sycophancy |