Intellectual Property Litigation, Protection, and AI Legal Matters: Defending the Algorithm™

Defending the Algorithm™ Newsletter: Edition 1

Welcome to the inaugural edition of Defending the Algorithm™, a LinkedIn newsletter dedicated to helping civil, IP and commercial litigation attorneys, insurance professionals, corporate counsel and all legal practitioners understand and navigate the rapidly evolving intersection of artificial intelligence and the law. In this series, we set forth a Bayesian Analysis of AI Litigation and Law. Thomas Bayes in the early 1700’s created his theorem of conditional probability and it is part of the framework for the predictive capabilities of AI and its LLM underpinnings. Trial lawyers use Bayesian reasoning every day.

This newsletter is a companion to our podcast series of the same name, which you can find

at this link: Defending the Algorithm™

The Defending the Algorithm™ podcast and blog series examines AI litigation trends, provides practical guidance on implementing AI tools in legal practice, and offers strategic insights for defending AI and IP related claims and lawsuits. Drawing on Bayesian probability analysis we analyze emerging case law and predict litigation trends across substantive law areas like intellectual property, trade secrets, employment, products liability and insurance coverage and bad faith litigation.

In this newsletter, we will bring you timely updates on landmark AI cases, practical implementation guidance, and strategic analysis that you can apply immediately in your practice of law and in client service. We will help you to create your lawyers AI Daily Routine and provide guidance on how to properly build and utilize your law firm’s tech stack. Corporate counsels are already asking law firms about their AI policies and are encouraging outside counsel to safely use AI for better and more efficient results in research and outcomes. AI in litigation is real and is not Y2K! We start with an important case update that we have written about extensively in our blog and podcast series. This newsletteer was authored and edited with assistance from Claude by Anthropic Opus 4.5 Max and Westlaw Precision AI.

CASE UPDATE: Lokken v. UnitedHealth Group

Discovery Battle Intensifies Over Alleged AI-Decided Healthcare Denials Case: Estate of Gene B. Lokken (Class), et al. v. UnitedHealth Group, Inc., et al.

Court: United States District Court, District of Minnesota

Case No.: 23-CV-03514 (JRT/SGE)

Status: Motion to Compel Discovery Filed by Lokken: Hearing Set for Feb. 11, 2026

Why This Case Matters to Insurance Carriers and Attorneys

The Lokken litigation represents one of the most significant AI accountability cases in the insurance industry. At its core, the case challenges whether UnitedHealth Group's alleged use of an AI algorithm called "nH Predict" to review post-acute care claims violates the insurance contracts with Medicare Advantage beneficiaries and whether that alleged conduct raises to the level of tortious behavior.

For those of us who follow AI (and/or insurance bad faith) litigation, this case exemplifies the "Defending the Algorithm™" challenge: when plaintiffs demand discovery and ESI production transparency from insurance carriers into algorithmic decision-making systems

- what must defendants produce, and how do courts balance the need for discovery against claims of burden, lack of proportion and the proprietary interests in AI systems?

The Allegations: AI Replacing Physician Judgment?

The plaintiffs—estates and individuals who were allegedly denied continued post-acute care coverage—allege that UnitedHealth systematically used the nH Predict algorithm to actually make medical necessity "recommendations" and claim denial decisions that were, in practice or reality, binding coverage denial decisions. According to the Amended Complaint:

• UnitedHealth's Evidence of Coverage ("EOC") documents promise that "Clinical Services Staff and Physicians make decisions on the health care services you receive based on the appropriateness of care and service and existence of coverage." nH Predict (as an LLM predictive model) was “trained” on massive quantities of patient treatment and outcome data – actual healthcare records of real patients.
• Plaintiffs allege on the contrary, that nH Predict generated actual claim denial recommendations, and employees who deviated from those recommendations faced discipline or termination.
• After UnitedHealth began using naviHealth and nH Predict in 2019, claim denial rates for post-acute care allegedly more than doubled—from 8.7% in 2019 to 22.7% in 2022
• Despite an 80% reversal rate on appeals—suggesting the initial denials were frequently erroneous—UnitedHealth allegedly continued relying on nH Predict's recommendations.

The plaintiffs do not merely challenge individual coverage decisions. They challenge the system itself—arguing that UnitedHealth's use of AI fundamentally breached its contractual and marketing promises to have humans make individualized medical necessity determinations. Defendant United Health denies these allegations and specifically denies that nH Predict makes claim decisions. They allege that all such decisions are made by well trained medical professionals.

The Discovery Dispute: Opening the "Black Box"

On January 28, 2026, plaintiffs filed their Memorandum of Law in Support of Motion to Compel Discovery Responses and Production of Documents. The motion reveals a fundamental tension in AI litigation: plaintiffs seeking transparency into algorithmic systems versus defendants resisting disclosure of what they characterize as burdensome discovery which seeks to invade proprietary technology and irrelevant information.

What Plaintiffs Want

The discovery requests at issue target the mathematical heart of the AI system. The Lokken

class plaintiffs seek production of, and the ability to forensically investigate and analyze:

1. The means, methods and data used in development, training and deployment of nH Predict AI — Documents related to how nH Predict was developed and trained, its structure [which would presumably involve production of algorithms and meta data], identities of individuals involved in creating the algorithm, design goals and anticipated benefits, and whether nH Predict was designed and trained as a predictive LLM to supplant physician decision-making. They also broadly seek production of Defendant’s policies, procedures, protocols, methods, practices, memoranda and reports concerning how they assess and adjudicate post-acute care claims; whether and how nH Predict complies with or adheres to generally accepted standards of medical practice; and Defendants coverage documents;
2. The naviHealth Acquisition — Documents about why UnitedHealth acquired naviHealth (inventor of nH Predict), projected cost-savings, and intended changes to naviHealth's practices.
3. Government Investigations — Documents concerning internal or governmental investigations into AI-driven denials, including documents previously produced to the U.S. Senate Permanent Subcommittee on Investigations (which had done some investigations of United Health).

• Employee Discipline and Performance — Performance metrics for employees who interface with nH Predict and disciplinary actions against employees who failed to follow algorithm recommendations, including documents related to employee training.
• Internal AI Oversight — Documents relating to UnitedHealth's "Artificial Intelligence

Review Board" and identities of members responsible for AI oversight.

Defendants' Position (They have not yet filed a brief in opposition)

According to the plaintiffs' brief, (and notes relating to their meet and confer sessions)

UnitedHealth has resisted this discovery on several grounds:

Merits-Based Objections: Defendants assert that "nH Predict is not used to make coverage determinations"—the very factual issue in dispute. Defendants therefore claim that there can be no discovery on the fundamental issue of whether or not nH Predict actually makes claim denials. Plaintiffs argue that this denial of liability cannot be used to limit discovery into the facts underlying that dispute.

Scope Objections: Defendants argue that Judge Tunheim's order on the motion to dismiss narrowed the contract claims. The Court has apparently rejected this interpretation twice.

Burden Objections: Defendants raised burden objections to nearly every request, but plaintiffs argue these objections are unsubstantiated by affidavits or supporting evidence as required.

Relevance Objections: Defendants contend that documents from before July 1, 2019 cannot be relevant. Plaintiffs counter that historical comparative data showing how denial rates changed over time with the acquisition of nH Predict is precisely relevant to proving their claims.

Key Legal Issues to Watch

1. Can Merits Denials Block Discovery? Plaintiffs argue that defendants cannot use their denial of plaintiffs' liability allegations as a basis to withhold discovery. This principle takes on special significance in AI cases where the information asymmetry is extreme.
2. The "Black Box" Problem in AI Discovery: Courts must balance plaintiffs' need to understand how proprietary AI systems work against defendants' interests in protecting trade secrets and proprietary algorithms. Further, even AI developers cannot explain how everything is working to make probability predictions inside the AI Black Box. The computer neural network, just like the human brain, cannot truly be mapped with certainty. This type of discovery, if allowed in this case and others, will be extensive and very expensive.

• Government Investigations as Discovery: Plaintiffs specifically seek documents already produced to the Senate Subcommittee. Those documents have already been searched, culled, and screened for privilege—so - producing them involves minimal additional burden to United Health or so argues the Plaintiff Class.

Applying the Bayesian Framework

In our Defending the Algorithm™ methodology, we use Bayesian conditional probability analysis to assess litigation risk and predict outcomes:

Prior Probability: Federal Courts generally favor broad discovery and have repeatedly rejected attempts to narrow discovery based on merits arguments.

Likelihood Function: The Lokken Court (per Judge Tunheim) has already denied defendants' attempt to bifurcate discovery (wherein they were implicitly seeking to avoid discovery of the algorithm and meta data) and rejected their reading of Judge Tunheim's order—twice.

Posterior Probability: Based on this evidence, we assess a HIGH probability that the Court will grant significant portions of the motion to compel, particularly for documents relating to nH Predict training and development, employee discipline, and government investigation materials.

What Happens Next

The hearing on the Motion to Compel is scheduled for February 11, 2026. As of this writing,

defendants have not yet filed an opposition brief.

We will continue to monitor this case and provide updates as the discovery dispute unfolds. The outcome will have significant implications for insurers in general using AI for underwriting or claims processing and/or for healthcare providers using AI in any form or at any time during the process of making coverage and claims decisions. Will Lokken create a national ESI standard for algorithmic discovery? Will other courts see these outcomes as deterministic for other cases – even if they are not actual precedent? Will attorneys be prepared to deal with these new challenges on both sides of the V. as they argue for, or against AI transparency litigation. More to come.

Connect With Us

For more analysis and practical guidance on AI and the law, visit our blog and podcast series: Click Here: Defending the Algorithm™ Blog and Podcast

Henry M. Sneath, Esq. Business, IP and AI Trial Lawyer Houston Harbaugh, P.C. Pittsburgh, Pennsylvania sneathhm@hh-law.com and 412-288-4013

This newsletter provides general information and does not constitute legal advice. The views expressed are those of the author and do not necessarily reflect the views of Houston Harbaugh, P.C. or its clients.

© 2026 Henry M. Sneath. All rights reserved. Defending the Algorithm™ is a mark used by Houston Harbaugh, P.C. for which registration application is in process with the USPTO.