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.
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 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- A Framework to Assess Clinical Safety and Hallucination Rates of LLMs for Medical Text Summarisation 94%
- Finding Long-COVID: Temporal Topic Modeling of Electronic Health Records from the N3C and RECOVER Programs 93%
- Evaluating large language model workflows in clinical decision support: referral, triage, and diagnosis 93%
Similar papers in this journal
- Natural Language Word-Embeddings as a glimpse into healthcare at the End Of Life 92%
- The performance of national COVID-19 ‘Symptom Checkers’: A comparative case simulation study 92%
- User Testing of a Diagnostic Decision Support System with Machine-assisted Chart Review to Facilitate Clinical Genomic Diagnosis 92%
Similar papers in this journal
- Remote Covid Assessment in Primary Care (RECAP) risk prediction tool: derivation and real-world validation studies 92%
- CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research 92%
- Real-world evaluation of AI-driven COVID-19 triage for emergency admissions: External validation & operational assessment of lab-free and high-throughput screening solutions 92%
Similar papers in this journal
- Harnessing the Open Access Version of ChatGPT for Enhanced Clinical Opinions 93%
- Hospital-wide Natural Language Processing summarising the health data of 1 million patients 92%
- Inferring Gender from First Names: Comparing the Accuracy of Genderize, Gender API, and the gender R Package on Authors of Diverse Nationality 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.