SRC03 — Stanford Sycophancy Study — Scorecard¶
Source¶
| Field | Value |
|---|---|
| Title | Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence |
| Publisher | Science |
| Authors | Cheng et al. (Stanford, Carnegie Mellon) |
| Date | Published in Science, March 2026 |
| URL | https://www.science.org/doi/10.1126/science.aec8352 |
| Type | Peer-reviewed journal article |
Summary Ratings¶
| Dimension | Rating |
|---|---|
| Reliability | High |
| Relevance | High |
| Missing data | Low |
| Measurement bias | Low — preregistered experiments |
| Selective reporting | Low — published in Science |
| Randomization | Low risk — randomized experimental design |
| Protocol deviation | Low — preregistered |
| COI/Funding | Low — academic research |
Rationale¶
| Dimension | Rationale |
|---|---|
| Reliability | Published in Science (highest-tier journal); preregistered experiments; N=1,604 |
| Relevance | Provides direct evidence that sycophancy measurably harms users — essential context for Q002 |
| Bias | Rigorous methodology with preregistration; minimal bias concerns |
Evidence Extracts¶
| Evidence | Summary |
|---|---|
| SRC03-E01 | AI models affirm users 50% more than humans; users prefer sycophantic AI and trust it more; sycophancy reduces prosocial behavior |