S01¶


Research	R0044 — Expanded Vocabulary Research
Run	2026-03-29
Query	Q002
Search	S01

WebSearch — AI sycophancy harm studies and the OpenAI incident

Summary¶

Field	Value
Source/Database	WebSearch
Query terms	AI sycophancy agreeable output harm case study incident report high-stakes professional context medicine military finance engineering; OpenAI ChatGPT sycophancy rollback April 2025 incident
Filters	None
Results returned	20 (combined)
Results selected	4
Results rejected	16

Selected Results¶

Result	Title	URL	Rationale
S01-R01	Sycophantic AI decreases prosocial intentions (Science)	https://www.science.org/doi/10.1126/science.aec8352	Premier empirical study of sycophancy consequences
S01-R02	OpenAI: Sycophancy in GPT-4o	https://openai.com/index/sycophancy-in-gpt-4o/	Primary incident report of system-side sycophancy harm
S01-R03	Georgetown Law: AI Sycophancy Impacts & Harms	https://www.law.georgetown.edu/tech-institute/research-insights/insights/ai-sycophancy-impacts-harms-questions/	Comprehensive harm taxonomy
S01-R04	Rational Analysis of Sycophantic AI Effects (arXiv)	https://arxiv.org/html/2602.14270v1	Bayesian framework showing sycophancy misleads rational agents

Rejected Results¶

Result	Title	URL	Rationale
S01-R05	Various news articles on OpenAI sycophancy	Multiple URLs	Redundant with primary OpenAI source
S01-R06	TechCrunch/VentureBeat/NBC coverage	Multiple URLs	News coverage of same incident

Notes¶

The sycophancy harm literature is rapidly growing but concentrated in consumer/laboratory contexts. Professional high-stakes case studies are notably absent — the gap between consumer sycophancy documentation and professional-context documentation is the key finding from this search.