Nippon Life v. OpenAI: Copyright Infringement in AI Training

Originally published in Spanish on derechoartificial.com. Adapted and contextualized for the US audience by Elena Markov.

Introduction

On March 4, 2026, Nippon Life Insurance Company of America filed a federal lawsuit against OpenAI in the Northern District of Illinois, alleging that the company's AI systems were trained on copyrighted insurance documents without authorization. The case, Nippon Life Insurance Company of America v. OpenAI Foundation and OpenAI Group PBC, raises fundamental questions about the boundaries of fair use doctrine in the age of generative artificial intelligence.

For US legal professionals, this litigation matters beyond its immediate parties. It represents the latest challenge to AI training practices from an industry — insurance — that maintains vast repositories of proprietary documents, actuarial data, and policy materials. How courts resolve the tension between copyright protection and AI development will shape the legal landscape for every organization that uses or develops machine learning systems.

Background of the Case

The Parties

Nippon Life Insurance Company of America (Nippon Life) is a subsidiary of Nippon Life Insurance Company, one of Japan's largest insurers. The company operates in the US market, providing disability and other insurance products to American employers and employees. In this capacity, it generates and maintains extensive documentation subject to copyright protection.

OpenAI Foundation and OpenAI Group PBC develop and operate ChatGPT, the generative AI system that has reshaped the technology landscape since its public launch.

The Alleged Infringement

According to the complaint, Nippon Life discovered that OpenAI's training datasets included proprietary insurance documents, actuarial tables, policy forms, and internal memoranda that the company had created and maintained under copyright protection. The documents at issue allegedly include:

• Disability insurance policy templates and rider forms
• Internal claims handling procedures and guidelines
• Actuarial modeling documentation
• Employee benefits plan materials
• Compliance and regulatory filings

Nippon Life alleges that OpenAI reproduced these materials during the training process without license, authorization, or compensation. The company claims that ChatGPT's outputs demonstrate knowledge of these proprietary documents — suggesting that the materials were incorporated into the AI's training corpus.

Procedural Posture

The complaint was filed on March 4, 2026, in the Northern District of Illinois. Nippon Life seeks declaratory and injunctive relief, statutory damages under 17 U.S.C. § 504, and actual damages for copyright infringement. The case is in its earliest stages; no responsive pleading has been filed as of this analysis.

Legal Framework: Copyright and AI Training

The US Approach: Fair Use Under 17 U.S.C. § 107

Copyright protection in the United States is governed by the Copyright Act of 1976, 17 U.S.C. §§ 101 et seq. Owners of copyrighted works hold exclusive rights to reproduce, distribute, and create derivative works based on their materials. These rights are subject to the fair use doctrine, which provides a defense for certain uses that serve purposes distinct from the original work.

Section 107 of the Copyright Act establishes a four-factor analysis for evaluating fair use:

Factor 1: Purpose and Character of the Use

Courts evaluate whether the new work serves a transformative purpose and whether it is commercial in nature. The Supreme Court's decision in Campbell v. Acuff-Rose Music, 510 U.S. 569 (1994), established that transformativeness — whether the new work "adds something new, with a further purpose or different character" — is central to this analysis.

Recent precedent suggests that mere reproduction for data analysis purposes may not qualify as transformative. In Authors Guild v. Google, 785 F.3d 51 (2d Cir. 2015), the Second Circuit held that Google's scanning of books for search and snippet display was transformative because it served a different purpose than the original publications. However, the court emphasized that Google did not reproduce entire works for end-user consumption.

The Supreme Court's 2023 decision in Andy Warhol Foundation v. Goldsmith, 598 U.S. 508 (2023), further narrowed the transformative use analysis. The Court held that the "purpose" of the secondary work must be distinct from the original — merely adding new expression or meaning does not satisfy the transformativeness standard. This decision has significant implications for AI training, as courts must now evaluate whether AI outputs serve a distinctly different purpose from the ingested materials.

Factor 2: Nature of the Copyrighted Work

This factor examines the type of work reproduced. Creative and expressive works receive stronger protection than factual compilations. Insurance documents occupy a middle ground: policy forms and procedural manuals are functional, but actuarial data and internal memoranda may contain creative elements.

Factor 3: Amount and Substantiality

Courts assess how much of the original work was reproduced and whether the portion taken constitutes the "heart" of the work. For AI training, the relevant question is whether training on a work — even if only excerpts appear in outputs — requires reproducing the work in its entirety. Courts have not definitively resolved whether the "amount used" analysis applies differently to training data than to direct reproduction.

Factor 4: Effect on the Market

The most consequential factor for AI training cases is market effect. This analysis examines whether the secondary use usurps the market for the original work or its derivatives. The Supreme Court clarified in Warhol that the relevant market includes works that serve as a "market substitute" for the original — not merely existing licensing markets.

For AI companies, the market harm analysis is complex. Training on insurance documents does not directly compete with Nippon Life's insurance products. However, if AI systems could eventually replace human insurance professionals, the market for creating proprietary insurance materials could be substantially impaired.

Court's Reasoning: Key Arguments

Nippon Life's Position

Nippon Life argues that OpenAI's training practices fail all four fair use factors:

On transformativeness: The company contends that OpenAI's use is not transformative because AI systems must ingest works in their entirety to learn from them. Unlike Google Books' snippet display, AI training requires reproducing complete works — the purpose and character of the use is replication, not transformation.

On market effect: Nippon Life argues that AI systems capable of answering questions about insurance documents create a market substitute for the original materials. Organizations could use ChatGPT instead of licensing or accessing Nippon Life's proprietary documentation.

On the "superseding use" doctrine: Building on Wheaton v. Peters, 37 U.S. (8 Pet.) 591 (1834), Nippon Life argues that AI outputs that allow users to understand insurance concepts without consulting the original works constitute a superseding use that harms the copyright owner's market.

The Broader Implications

The Nippon Life case differs from prior AI copyright litigation in its focus on insurance industry materials. While New York Times Co. v. Microsoft and Authors Guild v. OpenAI have addressed news articles and literary works, insurance documents represent a distinct category: functional materials that blend factual content with proprietary formatting, methodology, and business judgment.

This distinction matters for the fair use analysis. Insurance documents may receive thinner copyright protection than creative works, but they also embody significant investment in data compilation and analytical methodology. The case may require courts to develop more granular standards for evaluating AI training on functional versus expressive works.

Implications for Legal Professionals

For AI Developers and Technology Counsel

The Nippon Life case reinforces the urgency of securing licensing agreements for training data. While the fair use doctrine may provide a defense, litigation risk is substantial, and the legal standards remain unsettled. Technology counsel should advise AI development teams to:

• Conduct copyright audits of training datasets
• Prioritize licensing agreements for high-value content categories
• Document the transformativeness of AI outputs relative to training materials
• Monitor developments in pending litigation for precedential guidance

For Corporate Counsel and Compliance Officers

Organizations that maintain proprietary document repositories should evaluate their exposure as potential training data. While direct control over AI training is limited, companies can:

• Implement technical measures to prevent web scraping of proprietary content
• Review terms of service for AI platforms that may access organizational data
• Consider copyright registration for key proprietary materials
• Monitor AI outputs for evidence that proprietary content was incorporated into training

For Litigation Practitioners

The fair use defense in AI training cases presents novel questions that will require careful factual development. Key areas for discovery include:

• Training dataset composition and sourcing
• AI system architecture and the relationship between training data and outputs
• User interactions that demonstrate AI knowledge of specific copyrighted works
• Any internal assessments of copyright risk during product development

Comparative Context: The EU Approach

The European Union has adopted a different regulatory framework for text and data mining (TDM) activities. The Digital Single Market Directive (DSM Directive), 2019/790, establishes a mandatory exception for TDM research purposes (Article 4) and permits Member States to adopt exceptions for commercial TDM (Article 4(3)), subject to rights holders' ability to reserve their works for commercial TDM.

Unlike the US fair use doctrine — which depends on case-by-case judicial analysis — the EU framework provides clearer rules that rights holders can opt out of through reservation. This approach has both advantages and limitations: it provides certainty but may limit the flexibility that courts have developed under § 107.

For US practitioners advising clients operating internationally, understanding both frameworks is essential. A practice that constitutes fair use in the United States may violate EU copyright if conducted in or directed at European markets without compliance with DSM Directive requirements.

About the Author

Elena Markov is a specialist in algorithmic decision systems and computational ethics. She advises civil society organizations on AI accountability frameworks and researches the intersection of artificial intelligence and legal liability. View full profile.

This analysis is based on publicly available court filings and does not constitute legal advice. For questions about AI copyright liability, consult qualified legal counsel.

Related Coverage