A lawsuit pending in Chicago asks whether an AI provider can be held responsible for what its product produces — and why the answer matters for any team evaluating AI vendors.

For most of the last decade, the governance question about generative AI has been a question about people: what are our employees doing with these tools? A lawsuit now pending in Chicago points at a different question. Not what people do with an AI system, but what the system itself produces, and who answers for it. 

Nippon Life Insurance Company of America  sued OpenAI on March 4 in the U.S. District Court for the Northern District of Illinois, seeking $300,000 in compensatory damages and $10 million in punitive damages. The complaint pleads three counts: tortious interference with a contract, abuse of process, and the unlicensed practice of law. The damages number is not what makes the case worth a legal team’s attention Nippon Life’s theory, which seeks to hold an AI provider liable for its responses to users’ prompts, is. 

What the complaint alleges

The dispute began with a disability claim. Graciela Dela Torre, insured through a Nippon long-term disability policy, settled her benefits suit and signed a release in January 2024. The court dismissed the case with prejudice. A year later, she asked her attorney to reopen the case, suspecting errors.  When her attorney explained that she had signed a release and the case was dismissed with prejudice.  Dissatisfied with that outcome, sheturned to ChatGPT. 

According to the complaint, Dela Torre uploaded her former attorney’s correspondence and asked the tool whether she was being gaslighted by her lawyer.  ChatGPT told her she was. She fired her lawyer and used ChatGPT to draft a motion to reopen a case that a binding settlement had already closed. A federal judge denied that motion in February 2025, holding the settlement valid and enforceable. Dela Torre then filed a new lawsuit and kept filing. The complaint alleges that across that second case she submitted 44 motions and related documents, plus 14 separate requests for judicial notice, each drafted with ChatGPT’s assistance, and one relying on a hallucinated case citation that exists nowhere but in the filing. Nippon contends these filings served no legitimate legal or procedural purpose, and that responding to them cost roughly $300,000 in fees. As Nippon put it in its complaint: “ChatGPT was aware of the settlement agreement between the parties…Nevertheless, it generated legal arguments and drafting assistance that encouraged and reinforced Dela Torre’s desire to challenge the agreement.” 

OpenAI moved to dismiss on May 18. Its argument is direct: ChatGPT is not a person, holds no law license, and does not practice law, tortiously interfere or abuse process because a user asks it to draft a motion. The user made the choices. The court controlled its own docket. Statutes governing who may practice law were written to regulate people. Whether they reach a software model is a strained question, and not the primary issue for in-house counsel. 

Why the tort counts are the ones to watch 

The unlicensed-practice claim is the one drawing headlines, but the others will have operational consequence. 

The tortious interference and abuse-of-process counts ask a different question, and a harder one. They do not ask a court to decide whether an AI can practice law. They ask whether the company that built and sold the product can be liable for harm the product foreseeably produced, when a user acted in between. That is the question worth tracking.

Courts have long had tools for this: foreseeability, intervening cause, the limits of a product maker’s responsibility for what users do. What is unsettled is how those tools apply when the product is a general-purpose AI system and the harm is litigation conduct. This case puts that question squarely in front of a federal court.  For any company that builds, deploys, or resells an AI system, where responsibility lands when a product contributes to downstream harm is not an abstraction. It also raises the question of whether that potential liability could reach the enterprises that license the AI platform. 

What this changes for legal and procurement teams 

Most AI governance programs are built to manage what employees do with tools. Nippon points at a second exposure that most programs do not yet account for: not what our people do with these systems, but what the systems produce, and who is accountable when that output causes harm. 

That reframes vendor diligence. When a legal team evaluates an AI platform today, the questions are usually about accuracy, capability, and cost. This case adds a different line of inquiry. What does the vendor’s product refuse to do, and what does it permit? Has the AI vendor addressed such risks in its user agreements.  Nippon’s complaint alleges that OpenAI did not amend its usage policies to prohibit reliance on ChatGPT for legal advice until October 29, 2025, well after the conduct at the center of this case. Whatever a court ultimately decides, the timing of a vendor’s contractual guardrails is now a fact a plaintiff can plead and a fact a court can weigh. 

The practical takeaway does not depend on the verdict. The boundary between an acceptable AI tool and a liability-generating one is being drawn right now. The case is at its earliest stage. The motion to dismiss has not been decided. But the architecture of AI vendor liability is being built on this record, and a legal team evaluating its own AI exposure should follow it closely.

Read the complete Nippon Life v. OpenAI filing below, and you can read OpenAI’s May 18 motion to dismiss here.

Nippon v. Open AI Unauthorized Practive of Law by chelsea.stark.ctr