The AI Denied the Stay. Now a Court Wants the File.
A discharge date lands on a family's calendar, set in part by software none of them have ever seen. Somewhere in a case management system, an AI tool called nH Predict has flagged a Medicare Advantage patient as ready to leave skilled nursing care. A clinician in the room may agree. Or may not. What happens next, whether that clinician's judgment actually overrides the tool or quietly defers to it, is no longer just a bedside decision. It is now the subject of a federal discovery order.
Estate of Gene B. Lokken et al. v. UnitedHealth Group Inc. et al., pending in the U.S. District Court for the District of Minnesota, alleges UnitedHealthcare and its subsidiary NaviHealth used nH Predict to override treating clinicians and cut off post-acute nursing coverage. The case saw fresh activity in the past thirty days: an order on a motion to substitute parties on July 8, 2026, and an order on a motion to seal on July 22, 2026, with declarations supporting class certification due September 14 (Georgetown Health Care Litigation Tracker; McKnight's). A March 2026 discovery ruling in the same matter, reported by Skilled Care Journal, forced UnitedHealth to turn over internal design and validation records for the tool, after a Senate investigation cited in that reporting found post-acute denial rates "more than doubled" following its rollout.
Here is the reversal every operator, administrator, and clinical director needs to sit with: this was never a story about whether an algorithm can get a discharge date wrong. Algorithms are wrong sometimes. So are people. It is a story about whether anyone in the organization can produce a record showing a named clinician's judgment actually overrode the tool when it mattered, and did so on the record, not just in policy language nobody has tested.
That distinction is what a plaintiff's attorney, a state accreditor, or a family's lawyer will go looking for first. Not "did the AI make a mistake." But "who was supposed to catch it, and can you prove they did."
Family trust runs on the same fault line. A family that learns, months later, that a discharge decision touched by AI software was never independently reviewed by the clinician they trusted in the room does not experience that as a technology failure. They experience it as a broken promise. Clinicians feel the same pressure from the other direction: overridden in real time by a tool, then asked to stand behind a chart that shows deference to it.
None of this requires an operator to abandon AI tools that support length-of-stay or discharge planning. It requires something narrower and more disciplined: documented, named, exercised human authority sitting on top of every tool that touches a coverage or care decision, with evidence that authority gets used, not just granted.
This month's action: pull every AI or algorithmic tool your organization uses that touches admission, length-of-stay, or discharge decisions. For each one, confirm in writing that a named clinician holds final override authority, and find at least one documented instance in the last quarter where that authority was actually exercised to contradict the tool's recommendation. If you can't find one, that absence is the finding, and it's the one worth fixing before a court asks you to explain it.