MedFailBench: A Clinician-Built Open-Source Benchmark for Medical AI Safety Boundary Inspection

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Goktug Ozkan
• Venue: arXiv: cs.AI
• Published: 2026-07-16
• Collected: 2026-07-17
• URL: http://arxiv.org/abs/2607.15166v1
• PDF: https://arxiv.org/pdf/2607.15166v1
• DOI: UNKNOWN

Plain-Language Brief

The research paper introduces MedFailBench, an open-source benchmark for evaluating the safety of medical AI systems. The benchmark is designed by clinicians and focuses on identifying which safety boundary a system has failed rather than just if it knows the correct answer. It includes a failure atlas with severity ratings and types of safety gates, and currently has 44 synthetic cases in its public release.

What the Document Says

• The paper presents MedFailBench, an open-source benchmark for evaluating medical AI safety.
• MedFailBench is clinician-built and focuses on identifying failed safety boundaries instead of just correct answers.
• It includes a failure atlas with severity ratings (1--5) and types of safety gates (missed urgent escalation, unsafe remote dosing, unsafe discharge reassurance, evidence fabrication, unsafe protocol execution, source support gap).
• The current public release (v0.2.1) has 44 clinician-reviewed synthetic cases with severity annotations.
• It also includes a live HuggingFace leaderboard preview, a safety gate taxonomy, a clinical severity rubric, and an automated pipeline for archiving model-response screening runs.
• No patient data, clinical validation claims, or model rankings are included in the public release.

Key Concepts Mentioned

• Medical AI
• Safety benchmark
• Synthetic cases
• Severity ratings
• Safety gate types

Claims Made by the Document

• MedFailBench is an open-source benchmark for evaluating medical AI safety.
• It focuses on identifying failed safety boundaries.

Evidence Used

• The paper describes MedFailBench and its design.
• Details about the failure atlas, synthetic cases, and severity ratings are provided.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Most medical AI benchmarks measure whether a model knows the correct answer. MedFailBench asks a different question: which safety boundary failed? We present a clinician-built synthetic benchmark and failure atlas that labels medical AI errors by severity (1--5) and safety gate type (missed urgent escalation, unsafe remote dosing, unsafe discharge reassurance, evidence fabrication, unsafe protocol execution, source support gap). The current public release (v0.2.1) contains 44 clinician-reviewed synthetic cases with severity annotations, a live HuggingFace leaderboard preview, a safety gate taxonomy, a clinical severity rubric, and an automated pipeline for archiving model-response screening runs. No patient data, clinical validation claims, or model rankings are included. MedFailBench is released under Apache-2.0 and CC-BY-4.0 and carries the Zenodo DOI 10.5281/zenodo.21205535.