OpenAI and Microsoft Face Major Copyright Lawsuit Over AI Training on Books and Journalism

A sweeping federal copyright battle has exposed how OpenAI acquired and used copyrighted books and journalism in developing ChatGPT, drawing scrutiny of internal corporate decisions and its partnership with Microsoft. Prominent authors and major publishers are challenging the companies’ reliance on protected material without permission or compensation, while new evidence raises questions about the commercial consequences for professional writing. With the U.S. government and leading publishing organizations advancing competing interpretations of copyright law, the case could help determine the future of literary ownership and authors’ rights in an increasingly automated marketplace.

In the early development of the language models that would eventually power ChatGPT, OpenAI acquired collections of books from Library Genesis, an unauthorized digital repository containing millions of copyrighted works. The books supplied an unusually valuable resource for training artificial intelligence, offering the sustained prose, complex organization, and accumulated knowledge produced by generations of writers and publishers. Their acquisition also created a legal problem that would follow the company through years of extraordinary commercial expansion. Internal communications disclosed in federal litigation show employees discussing the risks associated with the material and efforts to remove it from company systems. By September 2026, those records had become part of a sweeping copyright dispute involving OpenAI, Microsoft and some of the most prominent writers and publishing organizations in the United States.¹

The allegations reach into the foundations of generative artificial intelligence. Authors contend that their books were acquired and reproduced without authorization, incorporated into commercial systems and subsequently used to develop products capable of competing with professionally written literature. OpenAI maintains that training language models involves a transformative use of existing material protected by American copyright law. Microsoft, whose investments and technical partnership helped support OpenAI's development, disputes allegations that its participation creates separate liability for infringement. The competing arguments are now before U.S. District Judge Sidney H. Stein in Manhattan, where lawsuits brought by authors and news publishers have been consolidated into one of the most consequential copyright proceedings involving artificial intelligence.²

On October 29, four influential publishing and copyright organizations intervened through a joint amicus curiae brief, urging the court to reject the companies' fair-use defenses. Their involvement follows the September filing of competing summary judgment motions and the disclosure of internal corporate statements concerning the commercial relationship between generative AI and the markets for books and journalism. The litigation has attracted the attention of the United States government, which has advanced its own interpretation of fair use, emphasizing technological competitiveness and national security. Together, the filings place a question of considerable economic significance before the federal judiciary, one whose resolution could influence the terms under which copyrighted literature is acquired, licensed, and used in the development of future technologies.³

The dispute began to take shape publicly in September 2023, when the Authors Guild and seventeen fiction writers filed a proposed class-action lawsuit against OpenAI. The plaintiffs included John Grisham, George R. R. Martin, Jodi Picoult, Jonathan Franzen and other authors whose works had established substantial commercial and literary reputations. Their complaint alleged that OpenAI had reproduced copyrighted books without permission to train the systems underlying ChatGPT. Microsoft was subsequently added as a defendant, while additional actions brought by nonfiction writers and other rights holders expanded the scope of the proceedings.

The New York Times filed its own lawsuit against OpenAI and Microsoft in December 2023, alleging that millions of its articles had been used without authorization in the development of artificial intelligence systems. Other news organizations pursued related claims involving copyrighted reporting, digital distribution and the ability of AI-generated responses to substitute for original journalism. In April 2025, the Judicial Panel on Multidistrict Litigation centralized the related federal actions before Judge Stein in the Southern District of New York. The resulting proceeding, In re OpenAI, Inc. Copyright Infringement Litigation, coordinates claims that originated in separate lawsuits and involve different categories of protected expression.⁴

The distinction between the literary and journalistic claims remains important. Books and news articles are produced under different commercial arrangements, and their markets can respond differently to technological competition. A novel may generate income through advances, royalties, foreign editions and adaptations over many years. A newspaper's reporting may depend upon subscriptions, advertising and the immediate attention of readers who visit its website. Both industries rely upon the legal rights attached to original expression, although the particular economic consequences of unauthorized reproduction may differ substantially.

The proceedings have developed around several forms of alleged infringement. The authors challenge the acquisition of books from unauthorized repositories, the copying required to prepare training datasets and the subsequent use of those works in model development. Some allegations concern transfers of protected material between OpenAI, Microsoft and third parties. Others address the content generated by ChatGPT, particularly where outputs allegedly reproduce recognizable elements of copyrighted fiction. The news publishers have advanced additional claims involving the reproduction of reporting and the potential displacement of traffic and revenue by AI-powered information services.

The underlying evidence has brought unusual attention to OpenAI's early acquisition practices. In its September 2026 summary judgment materials, the company acknowledged that compilations of books obtained from Library Genesis, commonly known as LibGen, were used to train GPT-3 in 2020 and GPT-3.5 in 2021. OpenAI maintains that other models implicated in the book-author proceedings were not trained on the same downloaded LibGen material. The admission establishes that books obtained from an unauthorized repository contributed to the development of important early language models. The legal consequences of that use remain contested.¹

LibGen has long operated outside conventional publishing channels, providing access to collections containing copyrighted fiction, nonfiction, textbooks, and scholarly publications. Its availability made enormous quantities of written material accessible without the negotiated agreements ordinarily required to reproduce or distribute commercially published books. For developers assembling training datasets, such collections offered a means of obtaining literature at a scale that would otherwise require extensive licensing negotiations with authors, publishers and other rights holders.

The commercial value of books in artificial intelligence development derives partly from qualities that distinguish published writing from much of the material gathered across the internet. Extended works contain coherent arguments, sustained narratives, specialized information and linguistic patterns refined through editorial processes. Those characteristics can help models learn relationships across lengthy passages and produce more organized responses. The technical value of the material has intensified the dispute over whether developers may incorporate it into commercial systems without compensating the parties responsible for creating it.

The authors' September summary judgment memorandum alleges that OpenAI obtained millions of books from unauthorized sources, reproduced protected works during model development, and participated in arrangements through which literary material was exchanged with Microsoft. Their legal arguments distinguish the initial acquisition of copyrighted files from their subsequent reproduction for training, maintaining that each activity must be evaluated according to the rights implicated and the circumstances in which it occurred.¹

OpenAI disputes the plaintiffs' characterization of its conduct. The company argues that large language models analyze written material to develop mathematical representations of linguistic relationships, enabling systems to perform functions substantially different from those served by the original publications. Its position emphasizes the purpose of training, the nature of the resulting technology and the absence of evidence that ordinary model use necessarily provides substitutes for the books incorporated into training datasets.

The internal records concerning LibGen have complicated that defense by drawing attention to decisions made after the material had been acquired. According to communications cited by the Authors Guild in September, OpenAI personnel discussed the legal and reputational consequences of retaining books obtained through the repository. The disclosed material includes exchanges involving researchers and executives who considered the risks associated with public knowledge of the company's acquisition practices. The plaintiffs also identified discussions about removing LibGen references and associated datasets from internal systems.⁵

A Slack channel connected to those efforts was renamed Project Clear, a detail that has attracted attention because of its proximity to conversations about deleting or removing the disputed material. The authors interpret the communications as evidence that company personnel recognized the legal difficulties surrounding their acquisition practices and attempted to limit their visibility. That interpretation is contested. Internal discussions about removing files, changing project names or managing reputational concerns do not independently establish deliberate concealment of infringement, and the legal significance of the underlying conduct remains unresolved.

The distinction became especially important during discovery proceedings involving OpenAI's Books1 and Books2 datasets. The parties disputed whether certain internal discussions concerning deletion and data management were protected by attorney-client privilege. Magistrate Judge Ona T. Wang considered whether communications reflected requests for legal advice or ordinary operational decisions, while OpenAI challenged aspects of the resulting discovery orders. In February 2026, Judge Stein set aside a prior ruling concerning privilege waiver over certain communications involving the deletion of those datasets and references to LibGen. The proceedings establish that the handling of the material became the subject of substantial discovery disputes, without establishing that the company unlawfully destroyed evidence.⁶

The documentary record nevertheless provides an account of how copyright concerns entered the technical and managerial decisions surrounding model development. Questions that might once have remained within engineering teams became matters for legal advisers, executives and federal courts. The authors' case depends partly upon the significance of those decisions and the extent to which they demonstrate knowledge of the rights attached to the materials being used.

By September 2026, the parties had accumulated enough evidence to seek summary judgment on major aspects of the litigation. The plaintiffs asked the court to find infringement and reject the defendants' fair-use defenses, while OpenAI and Microsoft sought rulings permitting their challenged activities under the Copyright Act. The motions were filed on September 4, with portions of the supporting materials subsequently made public after redactions were reconsidered. The court has not issued a final merits determination resolving the principal fair-use dispute as of October 29.²

The doctrine of fair use permits certain unauthorized uses of copyrighted material after consideration of four statutory factors. Courts examine the purpose and character of the challenged use, the nature of the protected work, the amount copied, and the potential effects upon the market for the original. These considerations have long governed disputes involving scholarship, criticism, technological innovation and other activities that incorporate existing expression. Their application to generative artificial intelligence has produced increasingly consequential litigation because training can involve copying entire works while the resulting systems perform functions far removed from conventional publication.

OpenAI maintains that its training practices are transformative. The company argues that the purpose of reproducing books during model development differs from the purpose of writing, publishing or reading those books. Its systems are designed to respond to prompts, assist with research, generate new text and perform tasks across numerous fields. According to its defense, training extracts statistical relationships and patterns from existing material, while the resulting models generally do not provide users with accessible copies of the original publications.

The authors reject the proposition that a different technological process necessarily establishes fair use. Their memorandum argues that OpenAI acquired complete copyrighted works for commercial purposes and developed systems capable of generating material that competes with those works. They emphasize the expressive character of literature, the substantial amount reproduced, and the emergence of commercial markets in which publishers already negotiate licenses for artificial intelligence training.¹

The existence of those licensing arrangements has become a central feature of the dispute. Authors and publishers argue that developers have opportunities to acquire material through negotiated agreements and that uncompensated copying diminishes the value of rights that copyright law ordinarily protects. OpenAI maintains that the fair-use analysis cannot be resolved simply by identifying a fee that rights holders might have charged, since lawful fair use necessarily permits certain activities without a license.

The dispute over compensation is inseparable from the structure of publishing contracts. Authors may retain certain subsidiary rights while granting publishers authority to reproduce, distribute, and license their work in defined formats or territories. Agreements negotiated before generative AI became commercially significant may not clearly allocate the right to authorize training. Publishers seeking to license extensive catalogs must therefore establish that they possess the relevant contractual authority, while individual writers may face considerable practical difficulties negotiating with technology companies on their own.

The economic stakes have grown as generative AI has developed beyond systems capable of answering questions into products used for drafting fiction, producing nonfiction, summarizing research and generating articles. The plaintiffs argue that such capabilities threaten the markets supporting human-authored literature, including the availability of commissions and readers' willingness to purchase professionally written work. The defendants maintain that broad predictions of future competition do not establish infringement or prove market harm caused by particular acts of copying.

The September filings brought additional scrutiny to statements attributed to technology-company executives concerning those markets. Reporting by Reuters and Bloomberg Law described previously redacted communications and testimony in which OpenAI and Microsoft personnel discussed the capacity of AI systems to substitute for published writing. OpenAI's head of ChatGPT, Nick Turley, was quoted as acknowledging that the company's products were substantially substitutive and could become more so as their capabilities improved. News publishers cited the remarks as evidence that the defendants understood the commercial consequences of their technology.⁷

The relevance of such statements depends upon their context and the legal questions under consideration. A corporate executive's observation that AI products can replace some uses of journalism may support an argument about commercial substitution. It does not establish that training on a particular protected work caused compensable injury, nor does it determine whether the challenged copying qualifies as fair use. The court must evaluate the statements alongside technical evidence, market analysis and the statutory factors governing the defense.

The authors have emphasized a related concern involving George R. R. Martin's A Song of Ice and Fire series. Their filings and the publishing organizations' October amicus brief reference efforts to develop language models capable of producing continuations of Martin's unfinished fictional universe. The plaintiffs argue that the capacity to generate material resembling forthcoming or existing literary works illustrates the possibility of direct competition with authors whose protected expression contributed to model training.¹,³

Copyright law draws distinctions among literary ideas, stylistic characteristics, and the particular expression embodied in a work. A system capable of producing text in a recognizable genre or style does not necessarily infringe the rights of an author associated with those characteristics. Outputs that reproduce substantial protected elements, including distinctive character development, narrative arrangements or expressive passages, may raise different questions. The legal analysis depends upon the content of the output and its relationship to the copyrighted work.

That distinction was central to a ruling issued by Judge Stein on October 27, 2025. OpenAI had sought dismissal of claims alleging that ChatGPT generated infringing outputs based on the authors' books. The court concluded that the consolidated complaint adequately alleged infringement concerning at least some outputs and permitted the claims to proceed. Stein's decision addressed the sufficiency of the allegations at the pleading stage. It did not establish that the disputed outputs were infringing or determine the ultimate liability of the defendants.⁸

The ruling nevertheless preserved an important avenue for the authors' case. The legality of using books to train a model and the legality of particular material generated by that model may require separate determinations. A court could potentially conclude that an underlying training use qualifies as fair use while finding that particular outputs reproduce protected expression without authorization. The same technology can therefore raise distinct legal questions at acquisition, training and generation.

The relationship between OpenAI and Microsoft introduces another layer of complexity. Microsoft's investments, computing infrastructure and commercial partnerships supported the development and deployment of OpenAI technology, while the companies pursued products incorporating related language models. The authors contend that the collaboration involved the transfer of copyrighted materials and provided Microsoft with sufficient financial benefit and supervisory authority to support secondary liability for alleged infringement.¹

The plaintiffs' allegations extend to arrangements through which the companies exchanged or supplied content used in model development. Newly disclosed materials described by Bloomberg Law refer to internal projects involving the collection and transfer of web content, including Project Taxi, associated with Microsoft's acquisition of extensive web material, and Project Mango, involving collection work performed for OpenAI. The plaintiffs argue that such arrangements reveal a degree of coordinated participation extending beyond an ordinary investment relationship.⁹

Microsoft disputes that the evidence establishes infringement or the legal conditions necessary for secondary liability. Its position emphasizes the transformative purpose of model development and challenges the contention that its technical support, commercial agreements or participation in product deployment automatically made it responsible for OpenAI's alleged violations.

The court will need to examine the particular acts attributed to Microsoft, the company's knowledge of the challenged conduct, its contractual authority, and the benefits allegedly derived from infringement. The legal responsibility of one corporate partner cannot be established solely through its financial association with another. The extent of Microsoft's involvement therefore remains a distinct question within the broader proceedings.

For news publishers, the consequences of generative AI have become especially visible through changes in how readers obtain information. The New York Times and other organizations allege that AI-powered services can provide answers derived from reporting without requiring users to visit the original publications. Such practices could reduce subscriptions, advertising opportunities and direct audience relationships, all of which contribute to the financial sustainability of professional journalism.

The publishers' September filings cite internal Microsoft data showing substantial differences between click-through rates associated with conventional Bing search and certain Copilot interactions. According to reporting on the unsealed materials, the figures indicated markedly lower referral rates for some publisher domains through AI-powered interfaces. The plaintiffs regard those results as evidence of commercial displacement, while the defendants dispute whether the data establish legally relevant market injury caused by the challenged copying.⁹

The distinction between reduced website traffic and copyright infringement requires careful examination. A change in reader behavior may have significant financial consequences without necessarily demonstrating unlawful reproduction of protected expression. Conversely, an AI-generated answer that supplies substantial copyrighted material or substitutes for the original publication could create a different legal claim. The court must determine which injuries are attributable to the alleged infringement and how they relate to the markets copyright law protects.

Evidence concerning ChatGPT's outputs has consequently become a major focus of discovery. News publishers sought access to large collections of conversation records to examine whether the system reproduced protected passages, how frequently relevant outputs appeared, and whether certain prompts could cause the model to generate material resembling original reporting. OpenAI resisted aspects of the requests, raising concerns about the sensitivity of user conversations and the burdens associated with production.

The dispute eventually produced orders requiring OpenAI to provide a sample of 20 million de-identified ChatGPT conversation logs. On January 5, 2026, Judge Stein upheld discovery orders issued by Magistrate Judge Wang, concluding that the production requirements appropriately addressed the evidentiary needs of the litigation while incorporating privacy protections. The orders required measures intended to remove identifying information and restrict the handling of confidential material.¹⁰

The volume of records involved illustrates the practical difficulty of establishing infringement in systems that generate responses across enormous numbers of interactions. A rights holder seeking to determine whether protected expression has been reproduced may need access to proprietary information concerning model behavior, training records or user outputs. Those materials may also contain commercially sensitive information and private communications belonging to individuals outside the litigation.

The proceedings have raised questions about how courts can provide meaningful discovery while protecting the privacy of users who had no role in the alleged infringement. Additional disputes in 2026 concerned the accessibility of records, the adequacy of OpenAI's searches and allegations that some information had been deleted or rendered unavailable. OpenAI denied accusations of improper evidence handling. Those disputes remain distinct from the underlying determination of whether the challenged uses of copyrighted material violated the law.

The difficulty of examining proprietary systems has become one of the defining features of AI copyright litigation. Authors and publishers may know that their works appeared in training collections without possessing complete information about the transformations, copies, and transfers that occurred during development. Establishing the relationship between a particular work and a model's commercial capabilities may require technical evidence unavailable outside the developer's internal systems.

In September, the United States government added an important institutional perspective. Its Statement of Interest, filed on September 1, urged the court to interpret fair use in a manner supportive of artificial intelligence development. The government emphasized national security, economic competitiveness and the benefits of maintaining a technologically capable domestic AI industry. It argued that training models on copyrighted material can serve purposes sufficiently different from those of the original works to qualify for protection under copyright law.¹¹

The filing reflected the administration's broader position that American leadership in artificial intelligence contributes to scientific progress, economic development and national security. The government warned that imposing extensive licensing obligations could raise the costs of developing advanced models, potentially favoring established firms with the resources to acquire large collections of protected material and creating disadvantages relative to foreign competitors operating under different legal regimes.

Its intervention did not resolve the disputed facts or establish that OpenAI and Microsoft had complied with copyright law. The government limited its principal analysis to training, recognizing that acquisition practices and particular outputs may present separate legal questions. It also acknowledged the possibility of statutory licensing arrangements or other mechanisms for compensating creators, while maintaining that such policy choices should not displace the existing fair-use analysis.¹¹

The government's position places the economic interests of copyright owners within a wider discussion of industrial policy. Licensing obligations might provide new sources of revenue for writers and publishers, while also increasing costs for companies developing models. The distribution of those costs could influence competition among technology firms, particularly smaller businesses unable to negotiate agreements comparable to those available to their largest competitors.

The publishing industry has advanced a different assessment of the same commercial circumstances. On October 29, the Association of American Publishers, Independent Book Publishers Association, International Association of Scientific, Technical & Medical Publishers and Copyright Clearance Center filed a joint amicus brief supporting the authors and news publishers. The organizations argued that commercial AI development should remain subject to established copyright protections and that the technological sophistication of a model does not independently justify uncompensated use of protected expression.³

Their submission emphasizes the commercial investment required to produce published works and the licensing markets through which intellectual property has long generated revenue. The organizations maintain that allowing developers to copy literature without permission could weaken those markets while enabling the production of competing content. The brief identifies existing licensing arrangements in book, educational, scientific, and technical publishing as evidence that rights holders are willing to negotiate authorized uses of their material.

The publishers also dispute the assumption that training necessarily transforms copyrighted writing into a use sufficiently different to qualify as fair use. Their argument places considerable weight on the expressive qualities of the material acquired and the capacity of generative systems to reproduce or commercially substitute for aspects of that expression. They contend that allowing unauthorized acquisition to displace negotiated licensing would shift economic value away from the authors and publishers responsible for producing the underlying works.³

The intervention carries significance beyond the individual plaintiffs because its participating organizations represent businesses operating under different publishing models. Scientific journals, professional reference publishers, independent presses and general trade publishers rely upon distinct combinations of subscription revenue, rights licensing, sales and institutional distribution. Their involvement reflects concern that a broad interpretation of fair use could influence numerous categories of published material.

American courts have already begun developing approaches to related disputes, although the decisions have reflected different factual records and legal theories. In Authors Guild v. Google, the U.S. Court of Appeals for the Second Circuit recognized fair use in Google's digitization of books for search and limited snippet display. The decision remains particularly relevant because it involved copying entire copyrighted works to provide a technologically distinct service without offering readers access to complete substitute editions.

OpenAI and Microsoft invoke similar reasoning in arguing that model training performs a transformative function. The authors distinguish generative AI from searchable book indexes, maintaining that language models can produce substantial new expressive content capable of competing with the works used in their development. The degree of market substitution, and the evidence necessary to establish it, remains a central point of disagreement.

Two decisions from California in June 2025 have also influenced the arguments. In Kadrey v. Meta Platforms, Judge Vince Chhabria found that Meta's training-related reproduction of thirteen authors' books qualified as fair use on the record presented. His opinion nevertheless recognized the potential for generative systems to undermine literary markets through the production of competing work. The authors had failed to develop sufficient evidence of that injury in their particular case.

In Bartz v. Anthropic, Judge William Alsup similarly recognized fair-use protection for certain training activities while distinguishing the acquisition and retention of pirated books from the subsequent development of a language model. The case later produced a $1.5 billion class-action settlement, illustrating the potentially substantial financial consequences of unauthorized book collections even when particular training uses receive legal protection.

Those decisions have helped clarify the importance of examining separate acts of acquisition, reproduction, retention and distribution. They have also revealed the limits of applying conclusions from one model, dataset or legal record to another. The outcome in Manhattan will depend upon the evidence concerning OpenAI and Microsoft and the legal precedents binding on the Southern District of New York.

The litigation additionally arrives amid broader judicial scrutiny of copyrighted material used to develop competing technologies. In Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence, the courts examined the use of protected Westlaw editorial material in an AI-assisted legal research product. The dispute has supplied another important reference point for arguments concerning commercial purpose, transformation and injury to existing markets. The technological and economic circumstances differ from those surrounding general-purpose language models, limiting the extent to which the cases can be treated as direct equivalents.

The unresolved questions carry substantial implications for the future of literary publishing. Authors and publishers rely on copyright not only to prevent unauthorized duplication but also to negotiate the terms under which their work is commercially exploited. Advances, royalties, foreign rights, adaptations and other licenses form part of an economic system that distributes revenue across the creation and publication of books. New technological uses may expand that system or alter its bargaining relationships, depending upon how the relevant rights are defined and enforced.

The development of generative AI has exposed uncertainties that traditional publishing contracts were rarely designed to resolve. An author may retain rights that a publisher has never previously had occasion to license. A publisher may control digital reproduction in one territory while another entity possesses authority elsewhere. Agreements governing training may require new language specifying permitted uses, compensation, retention of material and restrictions on onward distribution.

The legal result in Manhattan could influence the commercial incentives governing those negotiations. A ruling recognizing infringement in particular training practices could strengthen the position of rights holders seeking compensation and encourage more extensive licensing arrangements. A broad fair-use determination could reduce the circumstances in which payment may be required for training, while leaving separate liability questions concerning unauthorized acquisition, distribution and generated outputs.

The impact upon authors would be unlikely to fall uniformly across the publishing industry. Commercial nonfiction, genre fiction, academic scholarship and literary fiction occupy different markets, with varying degrees of dependence upon subscription services, library acquisitions, retail sales and subsidiary rights. The availability of AI-generated alternatives might affect some categories more readily than others, and the economic consequences could develop gradually as reader behavior and publishing practices change.

The international implications are considerable, although American judicial decisions do not determine the copyright laws of other countries. European legal frameworks include provisions concerning text and data mining, along with rights reservations and restrictions applicable under particular circumstances. Other jurisdictions have adopted different approaches to automated analysis and reproduction. Multinational technology companies must therefore operate across markets in which the same training practice may carry different legal consequences.

For international publishers, the prospect of AI licensing introduces practical questions involving territorial rights, collective management, the authority to sublicense protected material and the allocation of revenue among authors and publishing partners. The outcome of American litigation may influence commercial negotiations and corporate policies abroad without displacing the laws governing those transactions.

The proceedings have also demonstrated how difficult it can be for copyright owners to establish the commercial effects of technologies that incorporate protected works without necessarily reproducing them in recognizable form. A model may use literary material to improve its capabilities while generating new writing whose relationship to any particular source is difficult to trace. A publisher may experience reduced traffic or sales while facing considerable difficulty attributing that decline to the copying of specific works.

Courts must distinguish economic predictions from demonstrated injury, while recognizing that technological developments may affect markets in ways that existing methods of measurement do not readily capture. The evidence presented in the OpenAI litigation includes internal corporate communications, technical descriptions, output testing, licensing arrangements, and claims about changing reader behavior. Its significance will depend upon the court's assessment of those materials under the Copyright Act.

As of October 29, 2026, the principal summary judgment motions remain unresolved. The amicus brief filed by the publishing organizations represents a further intervention in an active dispute, and the remaining briefing and judicial consideration may refine the issues before any final ruling. No determination has established that all of OpenAI's challenged training practices infringed copyright, nor has the court accepted the companies' position that the conduct was protected by fair use.

The dispute has already brought previously inaccessible information about artificial intelligence development into the public record. The acquisition of books from unauthorized repositories, the internal discussions concerning their removal, the exchanges of material among commercial partners and the efforts to measure what language models reproduce have become subjects of judicial examination. The proceedings provide an unusually detailed account of the relationship between the creation of literary works and their subsequent use in computational systems.

For authors, that relationship concerns the practical value of rights acquired through years of professional work. For publishers, it reaches the commercial arrangements that support editorial investment and the development of new writing. OpenAI and Microsoft maintain that their technologies serve purposes sufficiently different from those of the source material to warrant fair-use protection. The plaintiffs contend that the companies have built valuable commercial products through unauthorized exploitation of protected works and have introduced competitors capable of diminishing the markets that support their creation.

The legal distinctions will matter long after the present litigation concludes. Reproducing a book to develop a model, distributing copies to another company, and generating an output that resembles protected fiction are different activities, with potentially different consequences under copyright law. The same underlying work can move through several of those processes, each involving distinct technical decisions and legal rights.

The books at the center of the proceedings were written for readers, acquired and published through established commercial arrangements and subsequently incorporated into technologies whose capabilities have altered the market for written expression. The courts must determine which uses fall within the rights reserved to authors, which may proceed without permission, and what evidence is necessary to establish infringement or market injury.

The decision will help define the conditions under which literary material can be used to develop commercial artificial intelligence. It will also influence the negotiations through which writers and publishers seek to retain economic participation in new uses of their work. The future value of a book may increasingly depend upon the legal treatment of activities occurring long after its publication, within systems whose technical operations remain largely beyond the author's control. The litigation in Manhattan has made those activities visible and placed their consequences before the courts.

References

1. Class plaintiffs. (2026, September 5). Redacted memorandum of law in support of motion for partial summary judgment, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143, ECF No. 1881 (S.D.N.Y.). Complete court filing, 49 pages.

2. OpenAI defendants. (2026, September). Memorandum of law in support of motion for summary judgment, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y.). Complete OpenAI summary judgment filing.

   Microsoft Corporation. (2026, September). Memorandum of law in support of motion for summary judgment, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y.). Complete Microsoft summary judgment filing.

3. Association of American Publishers, Independent Book Publishers Association, International Association of Scientific, Technical & Medical Publishers, & Copyright Clearance Center. (2026, October 29). Amicus curiae brief in support of plaintiffs, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y.). Complete joint amicus brief, 25 pages.

4. Authors Guild. (2026, September 5). Authors Guild and co-plaintiffs file motion for summary judgment v. OpenAI and Microsoft. Litigation background and filing announcement.

5. Authors Guild. (2026, September 21). Unsealed briefs in authors' case v. Microsoft/OpenAI reveal internal communications concerning copyrighted books. Research account and underlying litigation materials.

6. In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y. 2026). Discovery and privilege proceedings involving Books1, Books2 and LibGen-related communications. Case-document index and procedural record.

7. Brittain, B. (2026, September 17). OpenAI, Microsoft executives' quotes on AI training threaten copyright defense, news outlets argue. Reuters. Independent reporting on unsealed corporate communications.

8. Authors Guild et al. v. OpenAI Inc. et al., No. 25-md-3143 (S.D.N.Y. October 27, 2025). Opinion and order denying motion to dismiss output-based infringement claims. Complete judicial opinion.

9. News plaintiffs. (2026, September). Combined memorandum of law in support of motion for summary judgment, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y.). Complete news publishers' filing, 92 pages.

10. In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (S.D.N.Y. January 5, 2026). Order affirming discovery requirements concerning ChatGPT conversation logs, ECF No. 1021. Complete judicial order.

11. United States Department of Justice. (2026, September 1). Statement of interest of the United States, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143, ECF No. 1682 (S.D.N.Y.). Complete government statement, 20 pages.