Federal Judge Orders UnitedHealth to Disclose nH Predict AI Denial Algorithm with Alleged 90% Error Rate (2026)

Incident Summary

On March 9, 2026, a federal judge ordered UnitedHealth Group to disclose documentation for its nH Predict AI algorithm — a system used to make healthcare coverage denial decisions at scale.^[1] The lawsuit alleged that the nH Predict system has a 90% error rate, based on the observation that approximately 90% of claims denied by the algorithm were subsequently reversed when patients appealed.^[2] The court order required UnitedHealth to disclose the composition of its AI review board, staff compensation structures that may create incentives for claim denials, and the algorithm’s decision criteria.^[3] The case represents a significant escalation of the legal proceedings that began with initial reporting on UnitedHealth’s AI-driven claim denial system (documented in INC-23-0017). The 90% reversal rate, if confirmed through disclosure, would indicate that the AI system is systematically denying valid healthcare claims — effectively using automation to create barriers to care that most patients cannot overcome, since the appeal process requires time, resources, and persistence that many patients lack, particularly when ill.^[4]

Key Facts

• Court order: Federal judge ordered disclosure of nH Predict algorithm documentation on March 9, 2026^[1]
• Alleged error rate: 90% of AI-denied claims reversed on appeal^[2]
• Disclosure scope: AI review board composition, staff compensation structures, and algorithm decision criteria^[3]
• Prior incident: Extends INC-23-0017 (UnitedHealth AI Claim Denial) with major new legal development^[1]
• Systemic impact: UnitedHealth is the largest US health insurer, covering tens of millions of Americans

Threat Patterns Involved

Primary: Decision Loop Automation — The nH Predict system creates an automated decision loop where AI denies claims at scale, shifting the burden to patients to appeal each denial individually. The 90% reversal rate suggests the system functions not as a legitimate medical review tool but as an automated barrier designed to reduce payouts through friction.

Secondary: Overreliance & Automation Bias — The deployment of an AI system with an alleged 90% error rate for consequential healthcare decisions demonstrates extreme overreliance on automation for decisions that directly affect patient health outcomes and access to care.

Significance

1. Court-ordered AI algorithm disclosure — The judicial order to disclose algorithm documentation, review board composition, and compensation structures establishes a legal precedent for forced transparency of AI decision systems used in consequential contexts
2. 90% error rate in healthcare AI — If confirmed, a 90% reversal rate for AI-denied claims would represent the highest documented error rate for any consequential AI decision system, with direct implications for patient health outcomes
3. Friction as feature, not bug — The pattern of deny-then-reverse-on-appeal suggests the AI system may be designed to exploit the fact that most patients lack the resources to appeal, making the denial rate a revenue mechanism rather than a medical assessment
4. Extends ongoing UnitedHealth AI accountability — This court order represents a significant escalation from INC-23-0017, demonstrating that judicial intervention can force transparency where voluntary disclosure and regulatory action have failed