A study of clinical decision-making systems found that the agreement process itself can erase evidence of racial and demographic bias. Several AI agents deliberate and vote on diagnoses or treatment recommendations. The system resolves any disagreement and returns a single answer. That answer looks authoritative. The trail of dissent, and the bias that shaped it, disappears. The researchers call this disparity laundering. A lone AI making an obvious error is easy to catch: the answer looks wrong, the reviewer questions it, the system flags it. A panel of AIs that voted and agreed looks trustworthy even when the vote compressed a skewed signal into a clean result. The human reviewing the output sees consensus. The design gave them nothing to question. The implication for product teams is direct. Routing a decision through a panel of agents and showing the reviewer only the final answer removes the information they need to spot unfair outcomes. Fairness review requires access to how each agent voted, what each weighted, and where the disagreement was. Showing a confident result without that record conceals bias.