A taxonomy of medical hallucination types and a clinician-validated error-injection pipeline that creates matched correct and error-injected responses are developed, finding that more specific rubrics better distinguish correct from hallucinated responses.
Abstract
Hallucinations can undermine clinician trust in LLMs, making it important that evaluation methods capture clinically relevant errors. Rubric-based evaluation has become the leading approach for assessing LLMs in medicine, but it is unclear whether rubric scores reflect such errors. We first study this in a controlled setting using MedHallu, finding that more specific rubrics better distinguish correct from hallucinated responses. To test this systematically, we develop a taxonomy of medical hallucination types and a clinician-validated error-injection pipeline that creates matched correct and error-injected responses. Across HealthBench, HealthBench Professional, and LiveMedBench, our clinically relevant hallucinations are missed by rubrics, often leaving scores unchanged. We find that rubrics are most effective when explicitly checking facts, and are less effective for additional or unexpected errors they do not anticipate. A preliminary retrieval-based factuality check recovers some of the rubric-blind errors, suggesting a complementary approach. These findings reveal systematic blind spots in current medical evaluation of LLMs and suggest that rubric scores alone are insufficient to establish clinical reliability, potentially undermining clinician trust and confidence in clinical deployment.
Retrieval-based evidence verification provides a reproducible and transparent approach for evaluating the reliability of AI-generated medical information, with direct relevance to digital health practice, evidence-based medicine, and medical informatics.
A fine-grained hallucination classification framework based on medical knowledge graphs is proposed and a three-tier risk stratification scheme that references FDA medical AI software risk classification standards is established, highlighting the unique characteristics of the medical domain.
An empirical benchmark for evaluating clinical triage systems that assesses explanation quality alongside decision outcomes, and provides a reproducible, checkpoint-based evaluation pipeline and outline a roadmap for bias stress-testing, hallucination mitigation, and open benchmark release.
S. Marimuthu, Patricia L. Mabry, HealthPartners.Com· 0 citations
This work proposes a five-step diagnostic protocol covering feature correlation, bootstrap stability, sparse versus dense ranking disagreement, intervention baselines, and cross-dataset evaluation as a minimum standard for sparse-neuron localization claims, and investigates prior work using the proposed approach.
Huseyin Cavus, Sebin Sabu, J. Spear et al.· arXiv.org· 0 citations
Hallucination detection is particularly important for medical language models, but repeated-sampling approaches are computationally expensive. A faster alternative is a single-pass uncertainty head that predicts hallucination risk from a frozen generator's internal signals. Existing uncertainty heads consume backbone-s...
This work introduces a two-stage keyword-perturbation method for hallucination detection and extends the same probabilistic framework to four hallucination regimes: knowledge deficit, wrong knowledge, context distraction, and unstable inference.
Xu-Han Tong, Hao-Yue Bai, Da-Wei Zhou et al.· 0 citations