Gene Lokken was 91 years old when he fractured his leg and ankle. UnitedHealthcare covered his rehabilitation for 19 days. Then an algorithm called nH Predict flagged him for discharge. His family paid out of pocket for nearly a year. He died.
Roughly 90% of patients who appeal nH Predict denials win. Roughly 0.2% of patients ever appeal. The algorithm was built on a database of six million patients. The math was not an accident.
What This Looks Like at Scale
Cigna's PxDx system reviewed and denied 300,000 claims in two months. Average physician review time per claim: 1.2 seconds. These are not edge cases. This is the operating model. AI reviews. AI recommends. The human signs off fast enough to satisfy the process requirement and too fast to satisfy any meaningful accountability standard.
What nH Predict Is
nH Predict is an AI system built by naviHealth, a company UnitedHealth acquired in 2020 for $2.5 billion. It reviews post-acute rehabilitation stays and recommends when to stop coverage. A United States Senate investigation found that UHC's claim denial rate for post-acute care more than doubled after they began using nH Predict in 2019. Employees were pressured to follow its recommendations over their own clinical judgment.
The Evidence of Coverage documents that patients received promised that clinical staff and physicians make claims decisions. Not algorithms.
What the Court Is Asking For
In March 2026, a federal court ordered UnitedHealth to produce seven categories of discovery — including documentation of how nH Predict works and whether it was designed to supplant physician decision-making. The authorization record the court is asking for either exists and shows something damaging, or it doesn't exist in a form that survives scrutiny. Either answer is the gap.
The court found that plaintiffs are entitled to discovery of documents regarding how nH Predict works, its development goals, and whether it was designed to supplant physician decision-making.
That sentence is the accountability question made into a legal order.
Where the Gap Lives
Every article covering this case asks whether AI denied claims wrongly. Nobody has asked the prior question: who authorized nH Predict to make this class of decision, when, under what review, and does that authorization record exist in a form a court can verify?
The Accountability Report Six documented the pattern across five incidents: documentation of intent is not evidence of operational control. The Lokken discovery order is the first time a court has made that argument into a legal demand. The insurer had to build the governance record before the litigation arrived. The litigation arrived. The record is what the court is now compelling them to produce.
Twenty-five states now require human oversight for AI prior authorization decisions. None require a durable record of who authorized the AI to make this class of decision in the first place. The oversight can occur, satisfy the regulation entirely, and leave no documentation that it happened, when, or under whose authority.
The review happened. The authorization record for who decided that review threshold was acceptable does not have to exist under any current standard.
What Accountability Requires Here
A court is now compelling production of exactly the document that no AI governance standard required to exist. That is not a litigation risk. That is a governance failure that litigation has made visible.
Accountable by Design means the authorization record is built before the algorithm is deployed, not reconstructed under subpoena after someone dies.
Where in your organization right now is a tool running ahead of the rule designed to govern it? That gap is what we are here to map.
Vordan publishes the Gap Alert when the intelligence warrants it. The Accountability Report publishes every Sunday. The doctrine is Accountable by Design.
Estate of Lokken v. UnitedHealth Group, No. 23-CV-3514 (D. Minn.), Order on Motion to Compel, March 9, 2026 Hunton Andrews Kurth -- Court Allows Discovery Into Insurer's Use of AI to Deny Claims | https://www.hunton.com/hunton-insurance-recovery-blog/court-allows-discovery-into-insurers-use-of-ai-to-deny-claims alignmt.ai
-- From $556M to 1.2 Seconds: The Healthcare AI Cases That Changed Everything in 2026 | https://www.alignmt.ai/post/from-556m-to-1-2-seconds-the-healthcare-ai-cases-that-changed-everything-in-2026
KFF -- AI in Prior Authorization and Claims Review | https://www.kff.org/patient-consumer-protections/regulation-of-ai-in-prior-authorization-and-claims-review-a-look-at-federal-and-state-consumer-protections
Swept AI -- Lokken Ruling: AI Claim Denials Now Discoverable in Bad-Faith Suits | https://www.swept.ai/post/lokken-ruling-ai-claim-denial-discovery-bad-faith