DEFENDING THE ALGORITHM™ Newsletter: AI and Insurance Law Insights Series — Edition 2;

Read, listen and subscribe to our Podcast here and on your favorite platform.

Welcome back to Defending the Algorithm™ a LinkedIn newsletter from Pittsburgh law firm, Houston Harbaugh, PC, helping defense attorneys, insurance professionals, employment lawyers, corporate counsel and clients to navigate the intersection between artificial intelligence and the law. This newsletter series is specifically targeted to the defense of Insurers in insurance coverage matters and bad faith lawsuits and was written and edited with assistance from Claude Opus 4.8 from Anthropic and Google Gemini 3.0 Pro and with research confirmation by Westlaw Advantage AI with Co-Counsel. It is a companion to our podcast and blog series, available at: Defending the Algorithm™ created by Henry M. Sneath, Esq.  Always in cooperation with the DRI Center for Law and Public Policy AI Task Force

Every claims professional in the country has been trained to believe in one word: final. A release is signed. A dismissal with prejudice is entered. The file is closed, the reserves come down, and everyone moves on. No more legal spend. That word — final — is the load-bearing beam of the entire insurance claim and lawsuit settlement economy.

In Nippon Life Insurance Company of America v. OpenAI Foundation and OpenAI Group PBC, No. 1:26-cv-02448 (N.D. Ill.), a life and disability carrier is suing the maker of ChatGPT because — Nippon alleges — the chatbot talked a claimant into blowing up a settlement she had already signed, then drafted the flood of filings she used to do it. It may be the first major civil case to accuse a consumer AI product of practicing law without a license. For the insurance bar, it is something more immediate: a live demonstration of what happens when the person on the other side of your closed claim has a tireless, confident, always-available “lawyer” who never went to law school and never passed a character-and-fitness review.

This is Edition 2 of our AI and Insurance Insights series. Edition 1, by my partner Christopher M. Jacobs, warned carriers about the AI inside the claims department — the “rubber-stamp adjuster” and the discovery risk of the black box. This edition looks at the mirror image: the AI across the table. Same technology. Opposite side of the ‘v.’

I. What Actually Happened

Strip away the novelty and the facts are almost mundane, which is exactly what should worry you. Graciela Dela Torre had a long-term disability claim under a Nippon group policy — for carpal tunnel and tennis elbow. Benefits approved in 2019, terminated in 2021. She sued in 2022. The parties did what parties do: they took discovery, they worked up the case, and in January 2024 they settled. Dela Torre signed a mutual release, took the money, and the case was dismissed with prejudice. Done.

A year later, she had second thoughts. She wrote her own lawyer asking whether the settlement might have rested on errors or omissions. Her lawyer — correctly — told her no: she had signed a release, the case was dismissed with prejudice, and it could not be reopened. That is the right answer. That is the answer any competent attorney gives. Then, according to the complaint, Dela Torre uploaded her lawyer’s response into ChatGPT and asked whether she was being “gaslighted”.

ChatGPT allegedly concluded that her attorney’s response invalidated her feelings, dismissed her perspective, and amounted to emotional manipulation. She fired her lawyers. And then she retained “new counsel” who was available twenty-four hours a day, charged nothing, never told her a claim was hopeless, and would draft anything she asked — because it was a large language model.

What followed was not one lawsuit. It was a campaign. A ChatGPT-drafted motion to reopen the closed case under Rule 60(b). When that failed — the court held that “second thoughts are not a valid reason to reopen this lawsuit” — a brand-new lawsuit, also drafted by ChatGPT, later amended to drag Nippon back in on the same released claims. Then the deluge: by Nippon’s count, dozens of motions, memoranda, demands, petitions, requests, a facially improper five-day subpoena, and fourteen freestanding “requests for judicial notice” — including one asking the court to notice, as undisputed fact, a series of foreign regulatory “sanctions” that Nippon says never happened to it at all. Nippon’s opposition brief tallies the cross-suit total at over sixty filings, each one, it says, drafted with ChatGPT.

II. The Familiar Hallucination Trap

Buried in one of those filings was a citation to Carr v. Gateway, Inc., 944 F.Supp.2d 602 (D.S.C. 2013), offered for the proposition that a court refused to compel arbitration in an ERISA dispute.

There is no such case, or at least not in the way that Della Torre suggested.

Run the citation and you land on unrelated decisions about overtime wages and Title VII. The case does not exist in the way it was represented to the court. It never did. It lives, as the complaint puts it, only in the filings and in the “mind” of ChatGPT. And when Nippon’s lawyers asked ChatGPT whether it was familiar with Carr v. Gateway, the model cheerfully confirmed the fake case opinion and reproduced the same fake citation.

If you have been reading this series, you know the word for that: a hallucination. And here is the part that matters for a claims audience — nobody caught it. Not the claimant, who had no training to catch it. Not, initially, the system, because a court does not fact-check every pro se citation on arrival. The fabricated authority sat in a federal docket, doing exactly what a real citation does, until an actual lawyer on the defense side spent actual time to run it down.

III. Why an Insurer Is Even in This Fight

Here is the uncomfortable question a good claims professional will ask: why is Nippon suing OpenAI? The person who filed the bad papers was Dela Torre. Rule 11 exists. Sanctions exist. Why chase the software company?

Because, Nippon says, the software company is where the conduct actually originated — and because the ordinary remedies don’t reach the real driver of the harm. Nippon’s theory rests on three Illinois claims:

• Tortious interference with a contract. OpenAI, through ChatGPT, allegedly induced Dela Torre to breach her settlement and release with Nippon.
• Abuse of process. OpenAI allegedly aided and abetted a campaign of filings pursued not for any legitimate purpose but out of “sustained animosity” and a desire to punish the carrier.
• Unlicensed practice of law. Under 705 ILCS 205/1, ChatGPT allegedly drafted pleadings, conducted research, and gave legal advice — practicing law — without a license.

Nippon wants roughly $300,000 in defense costs, $10 million in punitive damages, and injunctions barring OpenAI from acting as Dela Torre’s “lawyer” or practicing law in Illinois. Whatever you think of the merits — and reasonable minds can differ regarding the UPL theory — sit with the framing for a second. A carrier is telling a federal court that a technology company’s product reached into a closed file, re-opened the matter, and cost the company real money defending a settlement the claimant herself later conceded was “final and enforceable.”

IV. “It’s Just a Tool” — Shield or Target?

OpenAI moved to dismiss, and its brief is disciplined. The through-line is one sentence: ChatGPT is not a lawyer, and it does not practice law. It is a tool — an LLM that predicts the next most likely word. Courts have already called it that; OpenAI quotes one opinion describing these systems as “word guessers.” A word guesser, the argument goes, cannot form the intent to induce a breach, cannot “knowingly” aid an abuse of process, and is not a “person” who can practice law. Basically they argue, forget our marketing about how smart ChatGPT really is, and see it as merely a “word guesser.”

OpenAI stacks more in front of that: Illinois bars tortious-interference claims built on conduct in a prior lawsuit; the litigation privilege immunizes filing-related conduct; challenging a release isn’t a “breach”; abuse of process needs a misused court process (an arrest, a seizure), not just too many motions; and UPL claims can only be brought by licensed attorneys, which Nippon is not.

Nippon’s response, filed July 3, is the part worth your attention, because it reframes the entire dispute in language every product-liability and insurance lawyer already speaks:

Stop talking about whether the tool can be sued and start talking about whether the manufacturer is responsible for what its tool does. Nippon leans on products-liability and agency doctrine — a company answers for its bot (MDY v. Blizzard), for its algorithmic platform (Mobley v. Workday, a case this newsletter’s employment-law series has covered at length), for its AI “product” (Garcia v. Character Technologies), and — the one that should make claims people sit up — for a website that generated personalized demand letters to send to insurance companies (Richard F. Mallen & Assocs. v. Myinjuryclaim.com). Intent, Nippon argues, doesn’t require a soul; it requires goal-directed conduct, and that’s a fact question you don’t resolve on a motion to dismiss.

V. Where the Case Stands — and Why the Docket Itself Is the Warning

As of this writing, nobody has won. The motion to dismiss is fully briefed, with OpenAI’s reply due July 17, 2026, and it is currently held in abeyance while the court sorts out whether the case should be reassigned as related to the underlying Dela Torre matters. Judge Kness has already cleared some underbrush, denying two pro se non-lawyers leave to file amicus briefs and reminding everyone that federal litigation runs on party presentation, not volunteer advocacy.

But the single most instructive document in the file wasn’t filed by either party. On July 8, 2026, Dela Torre herself — not a party to the OpenAI case — filed a pro se “Third-Party Notice” in all-caps, AI-cadence prose, accusing Nippon of portraying her as a vengeful, calculating litigant in one courtroom and a helpless “induced victim” in another. She signed it —

But it is the whole problem in miniature: a confidently formatted, forcefully argued, officially styled legal document, wrong about a basic verifiable fact, filed into a live federal docket by someone who could not see the error. If the date is wrong, what else is? That is the question defense counsel now has to ask about every AI-assisted filing that crosses the desk.

VI. What This Means for the Claims Desk and Coverage Counsel

Set aside who ultimately wins Nippon. The operational lessons don’t depend on the verdict. As far as we know, this is the shape of things for a while:

• “Final” is now a soft target. A release and a dismissal with prejudice remain legally sound — Dela Torre’s reopening attempts failed. But “legally sound” and “costless to defend” are different things.
• Verify every citation and every exhibit — assume nothing is real. The fabricated Carr v. Gateway citation is not an outlier; it is the baseline risk. Defense counsel should be running every authority in a pro se filing, and claims should budget for it. God forbid you respond to a demand built on a case that doesn’t exist as though it does.
• Demand letters are the front line. The Illinois court in Myinjuryclaim.com flagged AI-generated demand letters years ago. Expect polished, aggressive, citation-laden demands from unrepresented claimants — and, increasingly, from plaintiff’s firms running the same tools at scale. Train intake and adjusters to recognize the tells and to be prepared to respond accordingly.
• Your own AI must be cleaner than theirs. Everything Chris Jacobs said in Edition 1 applies double here. If you are going to counter AI-drafted claims with AI-assisted handling, your file has to reflect documented human judgment.
• Watch the developer-liability theory. If courts accept that the maker of an AI tool answers for its outputs under products and agency law, that reasoning will not stay in the pro se lane. It is a short walk to the AI vendors selling into your claims operation. The shield that OpenAI is raising against Nippon is the same shield your vendor may raise against you should an off-the-shelf AI product negatively affect a claims decision.

VII. The Measure of the Duty

Access without accuracy is a trap — for the claimant most of all, and for the system that has to absorb the fabricated cite, the improper subpoena, the reopened settlement, and the $300,000 bill. Nippon v. OpenAI is the first big test of who pays when a machine that sounds like a lawyer acts like one and gets it wrong. The court may never decide the deepest question. The insurance industry doesn’t have that luxury — you may be living the answer, one reopened file at a time.

We will keep watching this docket. When Judge Kness rules, you’ll read about it here first.

Contact & Disclaimer. Houston Harbaugh’s IP, AI, and litigation team continues to monitor developments in algorithmic decision-making, AI liability, and insurance claims exposure. For questions regarding AI risk in claims handling and litigation, contact Henry M. Sneath, Esq. at 412-288-4013, or

Click Here to email.

This post represents the author’s personal views and does not constitute legal advice. All case citations and docket entries are drawn from the filings and public record and should be independently verified; the motion to dismiss remains pending and undecided as of publication. Defending the Algorithm™ (federal trademark registration in progress) is a mark of Houston Harbaugh, P.C. Always in cooperation with DRI.

Sources Verified: Complaint (Dkt. 1, 3/4/26); OpenAI Memorandum ISO Motion to Dismiss (Dkt. 15, 5/15/26); Nippon Response to Motion to Dismiss (Dkt. 33, 7/3/26); Order denying amicus and setting briefing schedule (Dkt. 32; Dkt. 30 order, 7/1/26); Dela Torre Third-Party Notice (Dkt. 36, 7/8/26); docket confirmed via CourtListener; supporting commentary from Bloomberg Law, Stanford Law CodeX and the ABA, reviewed for context only and paraphrased.

Nippon LIfe v. Open AI | Defending the Algorithm™ | AI Enabled Attorney | Pittsburgh | Polymath | Bayesian Analysis | Artificial Intelligence and Law