OpenAI publishers copyright lawsuit: legal battle

What happened

OpenAI received formal service on August 27, 2026, in a new copyright infringement lawsuit filed by wikiHow, Inc. [3]. The court confirmed service for nine separate OpenAI entities on September 8, 2026, including OpenAI, Inc., OpenAI LLC, and OpenAI OpCo, LLC [3]. Service was accepted by agent Tatiyana Smith [3]. OpenAI must file its answer to the wikiHow complaint by September 17, 2026 [3].

On September 4, 2026, The New York Times Company, Daily News LP, and Ziff Davis Inc. filed redacted documents related to a motion for summary judgment in the consolidated proceedings [3]. This followed a September 3, 2026, omnibus sealing order issued by Judge Sidney H. Stein to govern summary judgment and Daubert briefing [3].

What changed vs. the baseline

The litigation has expanded beyond news and media publishers to include instructional content platforms. The wikiHow filing on August 21, 2026, introduced a new plaintiff into the SDNY proceedings under cause 17:101 [3].

Several items in the record are not new developments:

• Discovery Disputes: OpenAI’s demand for reporters’ notes and interview memos dates back to July 29, 2024 [1]. The Times’ opposition to these requests as "harassing" is a prior record [1].
• Oral Arguments: The order setting oral arguments for motions to dismiss and strike for October 8, 2025, is a historical event from the previous year [2].
• DOJ Intervention: The Department of Justice’s statement of interest supporting OpenAI was filed on September 2, 2026, and was already established in the baseline [3][6].

Why it matters

The entry of wikiHow indicates that OpenAI faces a broadening front of copyright claims. While news publishers focus on the "investment of enormous amount of time" in reporting, instructional platforms like wikiHow represent a different category of "how-to" data used for LLM training [1][3].

The September 3 sealing order and subsequent redacted filings signal that the parties are moving into the summary judgment phase [3]. This stage will force the court to address the DOJ’s recent argument that LLM training is a "transformative" fair use essential to national security against the publishers' claims of "stolen" content [3][6][7]. The outcome of the discovery dispute regarding reporter notes will set a precedent for whether AI defendants can probe the human authorship and "veracity" of the works they are accused of infringing [1].

Open questions

It is currently unconfirmed how OpenAI will respond to the wikiHow complaint by the September 17 deadline—specifically whether it will move to consolidate this case immediately with the existing multidistrict litigation (MDL) [3].

The court has yet to rule on the specific discovery motion regarding the production of reporters' notes [1]. Watch for a judicial order on this matter, as it will determine the scope of evidence plaintiffs must provide to prove their works are "authored by humans" and not already in the public domain [1]. Finally, the impact of the DOJ’s intervention on Judge Stein’s eventual summary judgment ruling remains the primary variable for the legal status of AI training data [3][6].

What happened

The New York Times filed a formal rebuttal in the Southern District of New York on September 4, 2026, directly challenging the U.S. Department of Justice’s (DOJ) intervention in its copyright lawsuit against OpenAI and Microsoft [1]. The Times alleged that the tech companies "stole" journalism to build commercial products that "substitute" for original reporting and "undermine" the news industry [1].

This filing follows a September 2, 2026, Statement of Interest from the DOJ, which urged Judge Sidney H. Stein to side with the AI developers [1]. The DOJ argued that restricting Large Language Model (LLM) development through a "misunderstanding of fair use doctrine" would damage American economic mobility and scientific progress [1].

The litigation, which began in 2023, now centers on two distinct phases of AI development:

• Input (Pre-training): The process where models ingest large-scale data to learn language patterns [1].
• Output (Post-training): The generation of human-like responses that the Times argues are often "substantially similar" to protected works [1].

OpenAI and Microsoft do not dispute using the publisher's content but maintain the practice is legally permissible under the fair use doctrine [1]. The Times is seeking monetary damages, a permanent injunction against further training on its material, and the "destruction" of any models already trained on its work [1].

What changed vs. the baseline

New Developments:

• Formal Rebuttal: The New York Times’ September 4 filing is a new response to the DOJ’s September 2 intervention [1]. It marks a shift from a bilateral dispute to a three-way legal argument involving the federal government.
• DOJ Policy Stance: While the baseline noted the DOJ's support for OpenAI, the full filing clarifies the government's specific fear: that a ruling for publishers would create a "barrier to entry" where only the largest firms could afford licensing, effectively stifling competition [1].

Prior Record (Not New):

• Attorney-Client Privilege Waiver: A ruling from November 2025 remains the standing precedent regarding evidence handling [2]. The court previously found that OpenAI waived privilege concerning the deletion of training datasets derived from "shadow libraries" like Library Genesis (LibGen) [2].
• Willfulness Defense: The court’s 2025 decision that OpenAI cannot invoke privilege while simultaneously claiming "good faith" or "innocent infringement" remains the operative discovery standard [2].
• Crime-Fraud Exception: The court's 2025 denial of the "crime-fraud exception"—finding no probable cause that communications were made to further a crime—remains unchanged [2].

Why it matters

The DOJ’s intervention signals that the executive branch views AI training as a matter of national economic policy rather than a narrow copyright dispute [1]. If the court adopts the DOJ’s view, it could "smooth the way" for universal AI training practices without the need for individual licenses [1].

Legal experts, including Harvard Law professor Lawrence Lessig, suggest that requiring rights clearance for all training data would be "disastrous" for innovation, as the current copyright system lacks an efficient way to identify and pay millions of individual owners [1]. Conversely, a ruling in favor of the Times could force a massive "destruction" of existing models, potentially resetting the technical progress of OpenAI’s GPT series and Microsoft’s Copilot [1].

The case is one of over 100 active copyright lawsuits against AI companies, making the SDNY’s eventual ruling the primary bellwether for the entire creative economy [1].

Open questions

• Substantial Similarity: While Stanford Law professor Mark Lemley notes that AI outputs only "sometimes" copy original works, the court has yet to define the specific threshold for what constitutes an infringing output in a commercial chatbot [1].
• Evidence of Deletion: It remains unclear what the compelled discovery regarding OpenAI’s 2022 internal communications will reveal about the "non-use" or deletion of pirated datasets [2].
• Sanctions: The court has not yet ruled on the July 2026 motion by publishers seeking sanctions against OpenAI for alleged evidence obstruction.
• Legislative Interference: With the DOJ citing "American prosperity," it is unclear if Congress will attempt to codify AI fair use before Judge Stein issues a final summary judgment.