{"type":"ScholarlyArticle","title":"Med-StepBench: A Hierarchical Reasoning Framework for Evaluating Hallucinations in Medical Vision-Language Models","authors":[{"name":"Minh Khoi Nguyen","slug":"minh-khoi-nguyen","given":"Minh Khoi","family":"Nguyen","affiliation":""},{"name":"Dai Lam Le","slug":"dai-lam-le","given":"Dai Lam","family":"Le","affiliation":""},{"name":"Amir Reza Jafari","slug":"amir-reza-jafari","given":"Amir Reza","family":"Jafari","affiliation":""},{"name":"Tuan Dung Nguyen","slug":"tuan-dung-nguyen","given":"Tuan Dung","family":"Nguyen","affiliation":""},{"name":"Hong Son Mai","slug":"hong-son-mai","given":"Hong Son","family":"Mai","affiliation":""},{"name":"Huy Thong Mai","slug":"huy-thong-mai","given":"Huy Thong","family":"Mai","affiliation":""},{"name":"Quang Huy Nguyen","slug":"quang-huy-nguyen","given":"Quang Huy","family":"Nguyen","affiliation":""},{"name":"Thanh Trung Nguyen","slug":"thanh-trung-nguyen","given":"Thanh Trung","family":"Nguyen","affiliation":""},{"name":"Reza Farahbakhsh","slug":"reza-farahbakhsh","given":"Reza","family":"Farahbakhsh","affiliation":""},{"name":"Noel Crespi","slug":"noel-crespi","given":"Noel","family":"Crespi","affiliation":""},{"name":"Phi Le Nguyen","slug":"phi-le-nguyen","given":"Phi Le","family":"Nguyen","affiliation":""}],"abstract":"Large vision-language models (VLMs) demonstrate strong performance in medical image understanding, but frequently generate clinically plausible yet incorrect statements, raising significant safety concerns. Existing medical hallucination benchmarks primarily focus on 2D imaging with one-shot diagnostic questions, offering limited insight into whether predictions are grounded in correct localization and abnormality identification, allowing critical reasoning errors to remain hidden behind seemingly correct diagnoses. We introduce Med-StepBench, the first large-scale benchmark for step-wise hallucination detection in 3D oncological PET/CT, comprising over 12,000 images and more than 1,000,000 image–statement pairs across volumetric and multi-view 2D data, which decomposes clinical reasoning into four expert-designed diagnostic stages. Using clinician-verified annotations, we perform the first step-level evaluation of general-purpose and medical VLMs, revealing systematic failure modes obscured by aggregate accuracy metrics. Furthermore, we show that current VLMs are highly susceptible to adversarial yet clinically plausible intermediate explanations, which significantly amplify hallucinations despite contradictory visual evidence. Together, our findings highlight fundamental limitations in grounding multi-step clinical reasoning and establish Med-StepBench as a rigorous benchmark for developing safer and more reliable medical VLMs.","doi":"10.24963/ijcai.2026/761","url":"https://ijcai.org/proceedings/2026/761","pdf":"https://ijcai.org/proceedings/2026/0761.pdf","markdown":"https://ijcai.org/proceedings/2026/761.md","bibtex":"https://ijcai.org/proceedings/2026/bibtex/761","volume":{"slug":"2026","title":"Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2026","url":"https://ijcai.org/proceedings/2026/","year":2026,"location":"Bremen, Germany","dates":"15-21 August 2026","editors":"Diego Calvanese","isbn":"978-1-956792-09-6","publisher":"International Joint Conferences on Artificial Intelligence Organization","doi_prefix":"10.24963/ijcai.2026"},"section":"AI and Health","track":"AI and Health","number":761,"pages":{"first":6841,"last":6849},"keywords":["Medical imaging: Medical imaging","Multimodal data: Multimodal data","Automated reasoning in clinical domains: Automated reasoning in clinical domains"],"videos":[],"references":[],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}