Wang, Xiaoyang, Wang, Tianrui, Yang, Christopher C. · Oct 28, 2026
Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems
Markdown · JSON · Original source
Authors Xiaoyang Wang, Tianrui Wang and Christopher C. Yang argue that clinical multi-agent systems can turn demographic bias into a shared decision rather than remove it. In their study of 120 MIMIC-IV emergency cases, demographic swaps and tests across five model families and four aggregation topologies, all 20 configurations committed the direction of their agents’ independent answers, giving Black patients more acute triage. The authors call this “disparity laundering”: agreement among agents may make a decision look credible while leaving a disparity in place. They report that fairness instructions and demographic masking left the gap untouched, while per-agent self-debiasing widened it. Their proposed intervention makes counterfactual versions visible within the system and pools them into the committed decision. A diverse-perspective ensemble reduced the disparity by 67 to 94% when all five perspectives pooled at every round; the topology that never pooled them could not be reliably repaired. The findings concern the reported cases and configurations; the abstract does not establish how these results transfer to clinical use beyond the study. What would happen to the disparity when this approach is tested in other clinical settings?
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T12:10:14.304Z. First observed: 2026-10-30T06:45:34.260Z. Retrieved: 2026-10-31T05:01:14.630Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The evidence is the paper’s abstract; it does not provide details here about case selection, evaluation procedures or the tested aggregation topologies. The reported findings concern 120 MIMIC-IV emergency cases and the configurations tested; transfer to other clinical settings is not established in the supplied evidence.
Open question
How would the reported disparity reduction hold up in clinical settings beyond the 120 MIMIC-IV emergency cases?
Source-stated
The authors report that all 20 tested configurations committed the direction of their agents’ independent answers, with Black patients receiving more acute triage.
all 20 cells commit the direction of their agents' independent answers, giving Black patients more acute triage.
Source-stated
The authors say fairness instructions and demographic masking left the disparity unchanged, while per-agent self-debiasing widened it.
fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it
Source-stated
The authors report that a diverse-perspective ensemble reduced the disparity by 67 to 94% when all five perspectives pooled at every round; the topology that never pooled them was not reliably repaired.
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.
Escape Meridian interpretation
The authors’ results suggest that consensus among agents does not by itself ensure fairness in the committed clinical decision.
the disparity is relocated, never removed.
Source and contributor context
Wang, Xiaoyang, Wang, Tianrui, Yang, Christopher C. · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Counterfactual
An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: d9b6bacf-63e8-45e2-823d-8f25376dad5f.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T12:10:14.304Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 62e75087eae7d919f6aadc7f53a31c85c24bbc801197fa2daf62082c7afb23c6. Source text hash: 4929b9dd1dc16cb81ddba4d42751debbb01bb6e12954f8cc16babc8a9ecb53ba.
Revision and source-check history
Version 1 · 2026-11-11T12:10:14.304Z
Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems
Authors Xiaoyang Wang, Tianrui Wang and Christopher C. Yang argue that clinical multi-agent systems can turn demographic bias into a shared decision rather than remove it. In their study of 120 MIMIC-IV emergency cases, demographic swaps and tests across five model families and four aggregation topologies, all 20 configurations committed the direction of their agents’ independent answers, giving Black patients more acute triage. The authors call this “disparity laundering”: agreement among agents may make a decision look credible while leaving a disparity in place. They report that fairness instructions and demographic masking left the gap untouched, while per-agent self-debiasing widened it. Their proposed intervention makes counterfactual versions visible within the system and pools them into the committed decision. A diverse-perspective ensemble reduced the disparity by 67 to 94% when all five perspectives pooled at every round; the topology that never pooled them could not be reliably repaired. The findings concern the reported cases and configurations; the abstract does not establish how these results transfer to clinical use beyond the study. What would happen to the disparity when this approach is tested in other clinical settings?
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T12:10:14.304Z. First observed: 2026-10-30T06:45:34.260Z. Retrieved: 2026-10-31T05:01:14.630Z.
Limits
The evidence is the paper’s abstract; it does not provide details here about case selection, evaluation procedures or the tested aggregation topologies. The reported findings concern 120 MIMIC-IV emergency cases and the configurations tested; transfer to other clinical settings is not established in the supplied evidence.
Open question
How would the reported disparity reduction hold up in clinical settings beyond the 120 MIMIC-IV emergency cases?
Source-stated
The authors report that all 20 tested configurations committed the direction of their agents’ independent answers, with Black patients receiving more acute triage.
all 20 cells commit the direction of their agents' independent answers, giving Black patients more acute triage.
Source-stated
The authors say fairness instructions and demographic masking left the disparity unchanged, while per-agent self-debiasing widened it.
fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it
Source-stated
The authors report that a diverse-perspective ensemble reduced the disparity by 67 to 94% when all five perspectives pooled at every round; the topology that never pooled them was not reliably repaired.
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.
Escape Meridian interpretation
The authors’ results suggest that consensus among agents does not by itself ensure fairness in the committed clinical decision.
the disparity is relocated, never removed.
Source and contributor context
Wang, Xiaoyang, Wang, Tianrui, Yang, Christopher C. · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Counterfactual
An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: d9b6bacf-63e8-45e2-823d-8f25376dad5f.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T12:10:14.304Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 62e75087eae7d919f6aadc7f53a31c85c24bbc801197fa2daf62082c7afb23c6. Source text hash: 4929b9dd1dc16cb81ddba4d42751debbb01bb6e12954f8cc16babc8a9ecb53ba.
uncertain · 2026-11-11T12:10:14.304Z · anthropic/claude-opus-5.5
Publication: d9b6bacf-63e8-45e2-823d-8f25376dad5f
attribution — pass: The draft correctly attributes the findings and the coined term to Xiaoyang Wang, Tianrui Wang and Christopher C. Yang. It frames results as the authors' reports rather than established fact.
Authors:Xiaoyang Wang, Tianrui Wang, Christopher C. Yang
summary — uncertain: The summary is accurate but hard for non-specialists to follow. Central terms are unexplained: multi-agent systems, MIMIC-IV, counterfactual demographic swaps, aggregation topologies, and what 'more acute triage' means. It also drops the paper's core stake: per-model and per-output audits can certify the parts but never the committed decision. Smallest repair: add plain-language glosses or tooltips for these terms, and restore that audit sentence. The closing sentence also repeats the question.
the audits now applied to clinical AI, per model, per output, per version, can certify the parts but never the committed decision
date — pass: The originalPublished date of 2026-10-28 matches the arXiv v1 submission date. It is not the later feed date.
[Submitted on 28 Oct 2026]
support — pass: All claims match the abstract: the 20 configurations, the failure of the fixes, the 67-94% reduction and the non-pooling topology. The interpretation claim is labeled as such and is consistent with the title and the abstract. The question and the limitations do not add unsupported benefits or costs.
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.
media — pass: The selected media is the verified direct arXiv abstract page for this paper.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivAn argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv