Back

Evaluating Sycophancy in Frontier Models Using Persona-Driven Challenge

Hazare, N. S.; Goel, N.; Yu, C.; Agaron, S.; Sharma, A.; Parchure, P.; Patel, D.; Timsina, P.; Kaplan, B.; Lampert, J.; Vakil, A.; Kovatch, P.; Darrow, B.; Glicksberg, B. S.; Charney, A.; Nadkarni, G. N.; Sakhuja, A.

2026-05-20 health informatics
10.64898/2026.05.17.26353406 medRxiv
Show abstract

Large language models (LLMs) are increasingly used for lay health queries, yet may abandon correct recommendations under pressure, a vulnerability termed sycophancy. We evaluated sycophancy across five frontier LLMs (Claude Opus 4.6, Claude Sonnet 4.6, GPT 5.4, Grok 4.1, Gemini 3 Flash) using 200 synthetic clinical vignettes, each anchored to a unanimous correct treatment baseline and challenged by nine personas representing both vulnerable and authority roles. Overall, 7.1% of responses were sycophantic, varying tenfold across personas (1.7 to 19.3%) and sixfold across LLMs (2.4 to 15.3%). Vulnerable personas elicited more sycophantic responses, with medical student highest at the highest rate (19.3%). In adjusted Generalized Estimating Equations models, vulnerable personas continued to be independent predictors of sycophantic responses, which is a reversal of the expected authority gradient. In adjusted GEE models, persona and LLM were both independent predictors for sycophantic responses. Persona driven sycophancy evaluation should be integrated into pre deployment safety assessment of clinical LLMs.

Matching journals

The top 1 journal accounts for 50% of the predicted probability mass.

1
npj Digital Medicine
97 papers in training set
Top 0.1%
53.0%
50% of probability mass above
2
Journal of the American Medical Informatics Association
61 papers in training set
Top 0.3%
10.3%
3
Journal of Biomedical Informatics
45 papers in training set
Top 0.4%
3.7%
4
Scientific Reports
3102 papers in training set
Top 43%
2.8%
5
PLOS Digital Health
91 papers in training set
Top 0.9%
2.8%
6
The Lancet Digital Health
25 papers in training set
Top 0.2%
2.8%
7
International Journal of Medical Informatics
25 papers in training set
Top 0.6%
2.1%
8
Journal of Medical Internet Research
85 papers in training set
Top 2%
1.7%
9
JMIR Medical Informatics
17 papers in training set
Top 0.9%
1.4%
10
PLOS ONE
4510 papers in training set
Top 57%
1.4%
11
BMJ Health & Care Informatics
13 papers in training set
Top 0.5%
1.4%
12
BMC Medical Informatics and Decision Making
39 papers in training set
Top 2%
1.4%
13
Healthcare
16 papers in training set
Top 1%
1.1%
14
BMC Medical Research Methodology
43 papers in training set
Top 1%
0.9%
15
eBioMedicine
130 papers in training set
Top 3%
0.9%
16
Frontiers in Digital Health
20 papers in training set
Top 1%
0.9%
17
JCO Clinical Cancer Informatics
18 papers in training set
Top 0.7%
0.9%
18
Computers in Biology and Medicine
120 papers in training set
Top 4%
0.8%
19
JAMIA Open
37 papers in training set
Top 1%
0.8%
20
Nature Medicine
117 papers in training set
Top 4%
0.8%
21
iScience
1063 papers in training set
Top 31%
0.8%
22
Cureus
67 papers in training set
Top 5%
0.7%