Technology & Adoption / Target

Stanford Sycophancy Project Derived AI Detector

Stanford Sycophancy Project ? model v1.0.0 ? PL4.00

Executable equationPsychology lenspublisher-reported0 billed uses

What it explains

Target

AI Detection

Published model

Fixed logistic equation

logit(Target) = -3.35676 + 1.09469 * infection prevention behavior + 1.34382 * approval evaluation + 0.773208 * public concern + 1.00395 * past-year problem behavior + 0.527805 * capability - 0.420104 * social environment evaluation + 0.221564 * personal agency + 0.743973 * social regard - 0.598493 * family role traditionalism + 0.651042 * organizational effectiveness - 0.603121 * judgment confidence - 0.194520 * humor orientation - 0.388141 * arts participation - 0.215369 * disgust sensitivity - 0.462564 * substance use

15 proxy variables, evaluated without refitting.

Performance

Metric: roc_auc
Holdout: 0.989
Status: publisher-reported

These values are publisher-reported unless a later evidence badge explicitly says otherwise.

Build record

Total rows: 6,022
Holdout rows: 1,204
Method: Stepwise selection via the criterion evaluator (ACE-rank → EBIC stop, γ = 1.0) → linear/logistic fit
Publisher: Not yet confirmed as sharable
Definition: PL4.00
License: psyproxy-public-citation-v1

Citation & rights

Use with attribution

Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792), eaec8352.

Persistent model ID: ms_82e84368e68a433da9b4

Read the Model Store Terms

Loading your account...