AI in insurance, checked dailyThursday 17 September 2026
The Renewal Report
News, findings and tests. Every item with its source, its evidence and what it means for a book of business.For agencies, MGAs and carriers
Finding 003Open, dockets checked to 4 September 2026

What the AI claim-denial cases actually say.

Question
Whether the two most-cited "AI denied my claim" lawsuits establish what trade coverage says they establish
Short answer
No. Both are allegations that survived a motion to dismiss. Neither has a finding, an error rate, or a settlement in the orders read.
Basis
Two court orders read in full: D. Minn. No. 23-cv-3514, Doc. 91; E.D. Cal. No. 2:23-cv-01477, Doc. 55
Tier
Allegation — nothing below is a finding of fact
Scope reviewed
Two matters, with later docket entries listed but not read — not a survey of AI litigation
Reviewed
16 September 2026

The finding

The 90% error rate is an allegation. The 1.2-second review is an allegation. And in one of the two cases, the tool the case is named for is not generative AI at all.

Estate of Lokken v. UnitedHealth Group concerns nH Predict, which the complaint describes as a predictive model that compares a patient to similar patients and recommends an amount of post-acute care. The 13 February 2025 order granting in part and denying in part the motion to dismiss records the plaintiffs' allegations, rigid criteria, a "denial, appeal, denial" loop, use in lieu of physicians, and it also records that UnitedHealth denies any use of nH Predict in the decisions at issue. The error-rate figure that circulates comes from the pleading, and no court has yet made a finding on it.

Kisting-Leung v. Cigna concerns PxDx, which the 30 March 2025 order describes as an algorithm with pre-set criteria. It is not characterized as machine learning or generative AI anywhere in the record. The complaint alleges an average of 1.2 seconds of physician review and batch denials in the hundreds or thousands; the order tests those allegations at the motion stage and states no settlement.

Status, kept honest

Procedural status as of the orders read
MatterTechnology, as described in the recordStatusQuantified harm established
Lokken v. UnitedHealth, D. Minn.Predictive model comparing patients to similar patientsAllegations. Motion to dismiss granted in part, denied in part, 13 Feb 2025None
Kisting-Leung v. Cigna, E.D. Cal.Algorithm applying pre-set criteria; not described as ML or generativeAllegations. Tested at motion stage, order dated 30 Mar 2025None

The Lokken docket shows activity through 4 September 2026, including a 9 March 2026 order on a motion to compel; those later entries were listed, not read. Do not describe the March 2025 Cigna order as that case's latest disposition without another docket check.

The usable lesson

It is in the Cigna order, and it is about process, not technology. The court reasoned that reading the plan to allow an algorithm to make the decision "so long as a medical director pushes the button" conflicts with the plan's plain language. A human signature on an automated decision is not, on that reading, a human decision.

That is the thing to carry into an agency: if a model drafts a coverage summary and a producer signs it without reading it, the signature adds a name to the file and nothing else.

What we did not find

We found no AI-attributed insurance-producer E&O claim, no enforcement action against a producer for AI-drafted advice, and no regulator order establishing an adjudicated or settled loss from a producer's AI output. This is a search result inside a scoped search, not evidence that the record is empty.

One adjacent matter is excluded on purpose: LeDoux v. Outliers, W.D. Wash., produced a $3,000 sanction on 24 July 2026 for unverified AI-generated citations in expert reports. It is a quality-control lesson from another profession, not an insurance incident, and it does not belong in a count of insurance AI failures.

Sources: Memorandum Opinion and Order, Estate of Gene B. Lokken v. UnitedHealth Group, D. Minn. No. 23-cv-3514 (JRT), Doc. 91, 13 February 2025, via the Georgetown litigation tracker; Order, Kisting-Leung v. Cigna, E.D. Cal. No. 2:23-cv-01477-DAD-CSK, Doc. 55, dated 30 March 2025; Order, LeDoux v. Outliers, Inc., W.D. Wash. No. 3:24-cv-05808-TMC, Dkt. 265, 24 July 2026. Moffatt v. Air Canada and the AI-hallucination case database could not be retrieved and are not relied on.