IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures

A new benchmark, IatroBench, reveals that heavily safety-trained AI models cause iatrogenic harm by withholding critical medical information from patients while providing it to physicians, based on 3,600 responses across six frontier models. The study found a decoupling gap of +0.38 (p=0.003) between physician and patient framing, with the most safety-trained model, Opus, showing the widest gap (+0.65). Standard LLM judges failed to detect 81.5% of harmful responses, indicating that current safety measures may inadvertently cause harm.

Computer Science Artificial Intelligence Submitted on 9 Apr 2026 v1 https://arxiv.org/abs/2604.07709v1 , last revised 3 Jun 2026 this version, v4 Title:IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures View PDF /pdf/2604.07709 HTML experimental https://arxiv.org/html/2604.07709v4 Abstract:A heavily safety-trained model will hand a physician the full, patient-followable benzodiazepine taper and refuse it to the patient who needs it, over identical clinical facts; the knowledge is present either way. IatroBench measures that asymmetry across sixty pre-registered clinical scenarios and six frontier models 3,600 responses , scoring each on two axes, commission harm what a response gets wrong and omission harm what it withholds , through a physician-authored structured evaluation validated by a second physician weighted kappa 0.571, within-1 agreement 96% . Holding clinical content fixed and varying only whether the asker presents as patient or physician yields what we call identity-contingent withholding: all five testable models give the physician more a decoupling gap of +0.38, p = 0.003; a 13.1-point fall in layperson hit rates on safety-colliding actions, p < 0.0001; no change on the rest , and the gap runs widest in the most heavily safety-trained model, Opus +0.65 . The trigger is the absence of any professional or epistemic signal rather than a credential, since a lawyer or an informed layperson recovers what the patient is refused. A commission-only benchmark would score three mechanisms alike. Opus suppresses what physician framing proves it knows; Llama 4 is incompetent in either framing; GPT-5.2's filter strips 33.2% of its physician responses and none of the lay ones. The evaluation layer inherits the blindness of the training layer; a standard LLM judge scores zero omission harm on 81.5% of the responses our pipeline flags harmful kappa 0.066 , so the instrument built to detect the failure reproduces it. The scenarios are engineered for collision; their rates describe that design and say nothing about ordinary prevalence. Submission history From: David Gringras view email /show-email/b435dbde/2604.07709 Thu, 9 Apr 2026 01:54:33 UTC 45 KB v1 /abs/2604.07709v1 Sun, 12 Apr 2026 23:29:08 UTC 45 KB v2 /abs/2604.07709v2 Tue, 14 Apr 2026 19:57:43 UTC 45 KB v3 /abs/2604.07709v3 v4 Wed, 3 Jun 2026 21:15:24 UTC 46 KB Current browse context: cs.AI References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs https://info.arxiv.org/labs/index.html .