Barrows v. Humana: the nH Predict AI coverage-denial suit where breach of contract, good faith and fraud survived a motion to dismiss

Verification status: CHECKING. The claims below are drawn only from sources fetched and archived this session. The ruling claim is quoted verbatim from the primary adjudicated record: the 14 August 2025 Memorandum Opinion & Order in Barrows v. Humana was text-extracted this session from the court PDF. This is an independent public-record account of an automated claims-denial dispute, not a client-outcome win: the subject is Humana and its plan members, and the court’s own words, not a vendor’s, are the record. The war-room never sets verified.

The problem

Medicare Advantage plans decide how long an insurer will pay for post-acute care: the skilled nursing and rehabilitation a patient needs to recover from a serious illness or injury (source). A putative class action pending in federal court in Kentucky alleges that Humana made those decisions with an AI model instead of a clinician: “Humana makes coverage determinations not based on individual patients’ needs, but based on the outputs of the nH Predict AI Model, resulting in the inappropriate denial of necessary care prescribed by the patients’ doctors” (source). An independent trade outlet framed the same allegation plainly: “A newly filed lawsuit alleges that Humana used artificial intelligence algorithms to deny key rehabilitation care for seniors” (source).

This is the case TIN’s verification thesis is built to examine: an AI step placed inside a consequential professional workflow, and an independent institution, here a federal court, forced to decide whether that workflow can be sued over at all.

What was built, and what is alleged

The model at the center of the suit is nH Predict, and the complaint ties it to the same tool named in the UnitedHealthcare litigation: “The plaintiffs say that the insurer relied on naviHealth’s nH Predict tool to make coverage determinations in long-term care” (source), and naviHealth “which created the nH Predict tool at the center of the suit, is owned by UnitedHealth Group’s Optum, and has been central to lawsuits against UHC” (source). The complaint alleges the model was paired with internal targets: employees were told to keep post-acute stays within 1% of the days the model projected, and “employees who deviate from the nH Predict AI Model projections are disciplined and terminated, regardless of whether a patient requires additional care” (source).

Humana disputes the characterisation. It “said it does not use artificial intelligence to make decisions about post-acute care coverage, but even if it did, the case is invalid due to the lack of jurisdiction” (source), and told the court the complaint “includes sensationalized allegations that Humana makes post-acute care coverage determinations based solely on the output of an Artificial Intelligence program,” adding that “Plantiff’s allegations are false” (source).

The outcome

The verifiable spine of this story is the court’s ruling on Humana’s motion to dismiss. On 14 August 2025, Judge Rebecca Grady Jennings issued the order, and its conclusion is explicit: the motion was “GRANTED in part and DENIED in part” (source). Four of the plaintiffs’ claims were allowed to proceed: “Count 1 (Breach of Contract), Count 2 (Breach of the Implied Covenant of Good Faith and Fair Dealing), Count 3 (Unjust Enrichment), and Count 8 (Common Law Fraud) may proceed” (source). The remaining counts did not survive: “Count 4 … Count 5 (Unfair Competition Law North Carolina), Count 6 (Insurance Bad Faith), and Count 7 (Unfair and Deceptive Insurance Practices) are DISMISSED with prejudice” (source), and the order records that plaintiffs conceded two of those counts were preempted by the Medicare Act (source). Two independent outlets reported the same count-by-count disposition. McKnight’s Long-Term Care News wrote that “Jennings let stand claims for breach of contract, breach of the good faith and fair dealing, unjust enrichment, and common law fraud,” and that “she did, however, dismiss four other claims regarding claims settlement, unfair competition, insurance bad faith and unfair and deceptive insurance practices because federal law preempted state law in those areas” (source). Bloomberg Law reported that the “decision by Judge Rebecca Grady Jennings … allows the affected class members to proceed with their claims that Humana engaged in fraud, unjust enrichment, breach of contract, and breach of good faith and fair dealing when it failed to disclose to beneficiaries that it had used AI software to make coverage determinations” (source). The ruling is a survival of the motion to dismiss, not a final judgment on the merits: the case proceeds.

The complaint’s headline figure is an allegation, and the story treats it as one. Pleaded “upon information and belief,” it asserts that “over 90 percent of patient claim denials are reversed through either an internal appeal process or through federal Administrative Law Judge (ALJ) proceedings,” which it argues “demonstrates the blatant inaccuracy of the nH Predict AI Model” (source). The same complaint alleges the appeal rate is tiny, that Humana relies on the model “because they know that only a tiny minority of policyholders (roughly 0.2%) will appeal denied claims” (source). Neither figure has been adjudicated or independently measured, and Humana denies the premise.

Weakest load-bearing source. The ruling claim rests on a Tier 1 primary, the 14 August 2025 Memorandum Opinion & Order, text-extracted this session from the archived court PDF, and is now corroborated by two independent Tier 2 legal outlets, McKnight’s Long-Term Care News and Bloomberg Law, each reporting the same count-by-count disposition and each fetched and archived this session. The weakest evidence in the story is therefore not the ruling but the complaint’s error-rate figures: the 90-percent-reversal and 0.2-percent-appeal figures are plaintiff allegations pleaded “upon information and belief,” not measurements, and are shown here as contested, with Humana’s denial quoted alongside them. They are not load-bearing for the story’s adjudicated core, which is what the court did with the counts. All sources are saved as sources/ copies and carry a web.archive.org snapshot recorded in dossier.json. No confirmation was sought from Humana; the badge never depends on a subject confirming its own record.

How this was verified. Method: every figure is quoted verbatim from a source fetched live this session (2026-09-01) and saved to sources/. The core allegation (Humana’s alleged use of nH Predict to deny post-acute care, the 1% internal target, the discipline of deviating employees, the 90%- and 0.2%- figures) is quoted from the primary complaint, Barrows et al. v. Humana, Inc., No. 3:23-cv-00654, Document 1, filed 12 December 2023 (Tier 1; local copy sources/complaint-doc1-2023-12-12.pdf, text sources/complaint-doc1-extracted.txt), and independently corroborated by Fierce Healthcare, “Class-action lawsuit accuses Humana of using AI algorithms to restrict access to care” (December 2023, Tier 2; local copy sources/fiercehealthcare-2023.html), which also carries Humana’s denial. The 14 August 2025 count-by-count disposition is quoted from the primary Memorandum Opinion & Order, Document 82, by District Judge Rebecca Grady Jennings (Tier 1; local copy sources/order-doc82-2025-08-15.pdf, text sources/order-doc82-extracted.txt), whose CONCLUSION was recovered by coordinate-ordered text extraction from the archived PDF, and independently corroborated by two Tier 2 legal outlets reporting the same count-by-count disposition: McKnight’s Long-Term Care News, “Humana must face class action suit over use of AI in denying post-acute care” (19 August 2025; local copy sources/mcknights-2025-ruling.txt), and Bloomberg Law, “AI Class Action Against Humana Gets Federal Judge’s Go-Ahead” (15 August 2025; local copy sources/bloomberglaw-2025-ruling.txt), both fetched live and archived this session. All sources carry a web.archive.org snapshot recorded in dossier.json. No confirmation was sought from any party. Checked 2026-09-01 (checking round).

Sources

1. U.S. District Court, W.D. Kentucky (Louisville Division) · Barrows et al. v. Humana, Inc., No. 3:23-cv-00654 — Class Action Complaint (Document 1) · 12 December 2023 · https://www.classaction.org/media/barrows-et-al-v-humana-inc.pdf — Tier 1 (the primary pleading; source of the core allegation, the 1% internal target, the discipline-of-employees allegation, and the 90%-reversal and 0.2%-appeal allegations pleaded “upon information and belief”; local copy sources/complaint-doc1-2023-12-12.pdf; Wayback http://web.archive.org/web/20260829154308/).
2. U.S. District Court, W.D. Kentucky (Louisville Division) · Barrows et al. v. Humana, Inc., No. 3:23-cv-654-RGJ, Document 82 — Memorandum Opinion & Order on Motion to Dismiss (District Judge Rebecca Grady Jennings) · signed 14 August 2025, filed 15 August 2025 · https://litigationtracker.law.georgetown.edu/wp-content/uploads/2023/12/Barrows-et-al_2025.08.15_MEMORANDUM-OPINION-ORDER.pdf — Tier 1 (the primary adjudicated record; fetched and text-extracted this session; source of the GRANTED-in-part / DENIED-in-part disposition, the four surviving counts and the four counts dismissed with prejudice; local copy sources/order-doc82-2025-08-15.pdf; Wayback http://web.archive.org/web/20260901214047/).
3. Fierce Healthcare · Class-action lawsuit accuses Humana of using AI algorithms to restrict access to care · December 2023 · https://www.fiercehealthcare.com/payers/class-action-lawsuit-accuses-humana-using-ai-algorithms-restrict-access-care — Tier 2 (independent trade press naming both parties; independent corroboration of the nH Predict allegation, the naviHealth/Optum provenance, and Humana’s denial that it uses AI to make coverage determinations and its “sensationalized” / “false” characterization; local copy sources/fiercehealthcare-2023.html; Wayback http://web.archive.org/web/20260901214119/).
4. McKnight’s Long-Term Care News · Humana must face class action suit over use of AI in denying post-acute care · 19 August 2025 · https://www.mcknights.com/news/humana-must-face-class-action-suit-over-use-of-ai-in-denying-post-acute-care/ — Tier 2 (independent trade press; independent corroboration of the 14 August 2025 count-by-count disposition, the four surviving counts and the four dismissed on preemption; fetched live this session; local copy sources/mcknights-2025-ruling.txt; Wayback http://web.archive.org/web/20260901220043/).
5. Bloomberg Law · AI Class Action Against Humana Gets Federal Judge’s Go-Ahead · 15 August 2025 · https://news.bloomberglaw.com/health-law-and-business/ai-class-action-against-humana-gets-federal-judges-go-ahead — Tier 2 (independent legal news; the disposition is in the free portion above the paywall; second independent corroboration of the four surviving counts; fetched live this session; local copy sources/bloomberglaw-2025-ruling.txt; Wayback http://web.archive.org/web/20260901215930/).

Related case files

• Lokken v. UnitedHealth: the sibling nH Predict suit, where the breach-of-contract claim survived Medicare preemption — the same AI model (nH Predict) and near-identical theory against the other largest Medicare Advantage insurer, useful for comparing how two federal courts handled the same automated coverage-denial claim.
• Cigna’s PXDX: an automated claims-review algorithm and the fiduciary-duty claim a federal court let proceed — a different insurer and system, same shape: bulk automated claim denial examined and partly survived in the courts rather than through the operator’s own account.
• Louis v. SafeRent: an algorithmic tenant-screening system that produced a $2.275m class settlement — a different sector, same pattern: an automated scoring system making high-stakes decisions about people, adjudicated against the public record.