aiqbench analysis
The study tests whether multi-agent clinical LLM teams reduce demographic bias in emergency triage. Across 120 MIMIC-IV cases with counterfactual demographic swaps, none of the 20 configurations removed the gap, and some topologies amplified it, with the worst case depending on the model. Late-stage fixes like fairness prompts and demographic masking did not help, and the authors report a diverse-perspective ensemble reduced the disparity by 67 to 94% in one setup, though it was not repaired in the topology that never pooled perspectives. For readers, the main takeaway is that per-model or per-output audits may miss bias that emerges at the level of the committed decision.
Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems
arXiv:2610.38215v1 Announce Type: new Abstract: Clinical large language models are increasingly fielded as multi-agent teams, on the assumption that deliberation averages out individual error. It does not average out bias. Across a full factorial of five model families and four aggregation topologies on 120 real MIMIC-IV emergency cases under counterfactual demographic swaps, no configuration neutralizes the disparity: all 20 cells commit the direction of their agents' independent answers, giving Black patients more acute triage. Every model has a topology that amplifies its baseline gap, but which one varies by model, so no design is reliably safe or reliably amplifying: the disparity is relocated, never removed. Late-stage fixes fail: fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it, because the bias is already encoded in the agents' ordinary clinical numbers. We call this disparity laundering; the audits now applied to clinical AI, per model, per output, per version, can certify the parts but never the committed decision. Interventions close the gap exactly when they make counterfactual versions visible inside the system and pool them, through deliberation, into the committed decision. A diverse-perspective ensemble does both while keeping the team, with no retraining or architectural change: the disparity falls by 67 to 94% where all five perspectives pool at every round, and the one topology that never pools them is the one it cannot reliably repair. Medical multi-agent systems do not filter demographic bias; they launder it into plausible decisions no single-version audit can catch, and the only defense is to make the bias visible.
Abstract:Clinical large language models are increasingly fielded as multi-agent teams, on the assumption that deliberation averages out individual error. It does not average out bias. Across a full factorial of five model families and four aggregation topologies on 120 real MIMIC-IV emergency cases under counterfactual demographic swaps, no configuration neutralizes the disparity: all 20 cells commit the direction of their agents' independent answers, giving Black patients more acute triage. Every model has a topology that amplifies its baseline gap, but which one varies by model, so no design is reliably safe or reliably amplifying: the disparity is relocated, never removed. Late-stage fixes fail: fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it, because the bias is already encoded in the agents' ordinary clinical numbers. We call this disparity laundering; the audits now applied to clinical AI, per model, per output, per version, can certify the parts but never the committed decision. Interventions close the gap exactly when they make counterfactual versions visible inside the system and pool them, through deliberation, into the committed decision. A diverse-perspective ensemble does both while keeping the team, with no retraining or architectural change: the disparity falls by 67 to 94% where all five perspectives pool at every round, and the one topology that never pools them is the one it cannot reliably repair. Medical multi-agent systems do not filter demographic bias; they launder it into plausible decisions no single-version audit can catch, and the only defense is to make the bias visible.
21 pages, 8 figures. Code: this https URL . Committed to EACL 2027
Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
arXiv:2610.38215 [cs.AI]
(or arXiv:2610.38215v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.38215 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 361 words
Related coverage
Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows
arXiv · Sep 17, 2026
They'll Verify. They Just Won't Act. How Authority Framing and Laundered Code Turn a Trusted Agentic CI/CD Pipeline Into an Attack Surface
arXiv · Jul 23, 2026
PCM: A Proof-Carrying Message Protocol for Verifiable Multi-Agent Communication
arXiv · Oct 31, 2026
Agentic AI Defense: Decode-Time Provenance Binding for MCP
arXiv · Oct 27, 2026