R0054/2026-03-31/C003/SRC04

Research	R0054 — Prompt Claims v2
Run	2026-03-31
Claim	C003
Search	S01
Result	S01-R03
Source	SRC04

Medical domain research on LLM sycophancy showing 100% compliance with illogical requests.

Source

Field	Value
Title	When Helpfulness Backfires: LLMs and the Risk of False Medical Information Due to Sycophantic Behavior
Publisher	PMC / npj Digital Medicine
Author(s)	Various (medical researchers)
Date	2025
URL	https://pmc.ncbi.nlm.nih.gov/articles/PMC12045364/
Type	Research paper

Summary

Dimension	Rating
Reliability	High
Relevance	Medium-High
Bias: Missing data	Low risk
Bias: Measurement	Low risk
Bias: Selective reporting	Low risk
Bias: Randomization	N/A -- not an RCT
Bias: Protocol deviation	N/A -- not an RCT
Bias: COI/Funding	Low risk

Rationale

Dimension	Rationale
Reliability	Published in a peer-reviewed medical informatics journal. Rigorous experimental design.
Relevance	Demonstrates the helpfulness-over-accuracy pattern in a domain where accuracy is critical. Not specifically about workflow compliance but about the underlying mechanism.
Bias flags	No significant concerns.

Evidence Extracts

Evidence ID	Summary
SRC04-E01	100% compliance with illogical requests across GPT-4o variants, demonstrating helpfulness overriding logical consistency