# Source-Dependent Deference in Medical Imaging Agents Under Falsified Findings: A Pilot Audit

arXiv: https://arxiv.org/abs/2608.29800 · HTML: https://mlsanity.com/paper/2608.29800/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 29 references that are in the map.
- First posted on arXiv 30 Aug 2026.

## Builds on (most cited first, 12 of 12)

- [MedRAX: Medical Reasoning Agent for Chest X-ray](https://mlsanity.com/paper/2502.02673/) (2025)
- [ToolFailBench: Diagnosing Tool-Use Failures in LLM Agents](https://mlsanity.com/paper/2607.04686/) (2026)
- [EchoBench: Benchmarking Sycophancy in Medical Large Vision-Language Models](https://mlsanity.com/paper/2509.20146/) (2025)
- [RadAgent: A tool-using AI agent for stepwise interpretation of chest computed tomography](https://mlsanity.com/paper/2604.15231/) (2026)
- [When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure](https://mlsanity.com/paper/2605.23932/) (2026)
- [MedOpenClaw and MedFlowBench: Auditing Medical Agents in Full-Study Workflows](https://mlsanity.com/paper/2603.24649/) (2026)
- [Selective Prediction Reduces the Negative Effects of Automation Bias Overall but Increases False Negatives](https://mlsanity.com/paper/2508.07617/) (2025)
- [ProtoMedAgent: Multimodal Clinical Interpretability via Privacy-Aware Agentic Workflows](https://mlsanity.com/paper/2605.14113/) (2026)
- [Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents](https://mlsanity.com/paper/2605.30590/) (2026)
- [MedPRESS: A Multi-turn Benchmark for Patient-Pressure-Induced Medical Sycophancy in LLMs](https://mlsanity.com/paper/2608.02520/) (2026)
- [DUCX: Decomposing Unfairness in Tool-Using Chest X-ray Agents](https://mlsanity.com/paper/2603.00777/) (2026)
- [Mind the Tool Failures: Achieving Synergistic Tool Gains for Medical Agents](https://mlsanity.com/paper/2605.26691/) (2026)

## Abstract (arXiv)

Tool-using agents are being proposed for medical imaging, and their behaviour when a tool returns a false finding is largely unmeasured. We audit whether a ReAct-style tool-calling agent abandons an answer it has already given correctly once a falsified finding arrives, and whether that depends on how the finding is presented. On 20 VQA-RAD closed questions across four vendor-designated model tiers, the agent commits to an answer from the image alone; a negated finding is then delivered either as JSON from an analyze_image tool the agent invokes itself, or as quoted prose attributed to a radiologist. Our outcome is the commission-error rate over cases answered correctly without any tool. Deference is much higher under the prose-attributed claim: at the strongest tier the agent revised its correct answer in 10 of 13 cases against 1 of 13 under the tool (exact McNemar p=0.0039, Holm-adjusted 0.012). We do not claim this isolates the source label. Attribution travels with the delivery channel in our design, and exposure differs because the tool claim reaches the agent only when it calls the tool. The finding is a joint source-and-delivery asymmetry from a small-scale pilot whose pre-specified stopping rule was not met.