[arXiv]score: 0.35

Measuring Narrative Anchoring Bias in Clinical Language Models

July 31, 2026

Clinical LLMs exhibit Narrative Anchoring, where identical medical facts result in different diagnoses depending on the sociolinguistic register used. A benchmark of 1,000 USMLE vignettes shows a Narrative Anchoring Gap of 0.064 across seven models without using explicit demographic markers.

HOW THIS AFFECTS YOU

●

researcherYou should test clinical models against sociolinguistic register variation to ensure diagnostic robustness.

●

policyThis reveals a new category of bias that functions independently of explicit demographic identity tokens.

●

healthYou must account for linguistic style when deploying diagnostic models to avoid inconsistent patient care.

read original ↗arxiv.org

DAILY DIGEST

you don't check 9 sources — we do. one email every morning, read in 2 min. free. unsubscribe anytime. privacy

← back to feed