New York Times v. Microsoft Corp.

In copyright actions brought by news organizations against OpenAI and Microsoft, district court allows direct and contributory copyright infringement and trademark dilution claims to proceed and dismisses DMCA and unfair competition claims.

In 2023, The New York Times, The New York Daily News and eight regional newspapers, along with the Center for Investigative Research Inc., filed lawsuits against AI research company OpenAI, OpenAI-related entities (collectively, OpenAI), and investor and partner Microsoft Corp. The news organizations asserted claims for direct and contributory copyright infringement, trademark dilution, violations of the Digital Millennium Copyright Act (DMCA) and unfair competition related to misappropriation. The news organizations alleged defendants train their large language models (LLMs) using content collected from the internet, including copyrighted content from the plaintiffs’ websites. The data is then used to inform LLM responses to user queries, which may result in the LLMs “regurgitating” large portions of the news organizations’ content.

Defendants moved to dismiss the contributory copyright infringement, DMCA and common law unfair competition by misappropriation claims. Microsoft moved to dismiss certain state law trade dilution claims, and OpenAI moved to dismiss certain direct copyright infringement and federal trademark dilution claims. 

On the direct infringement claims, OpenAI argued that the Copyright Act bars claims involving conduct occurring more than three years prior to the filing of the complaint. The district court disagreed, holding that, although the news organizations alleged defendants trained their LLMs in 2019 and 2020, the complaints did not establish that they discovered or with due diligence should have discovered that fact in 2019 and 2020. In its reasoning, the district court emphasized the fact that minimal articles discussing OpenAI’s products existed at the time, and—regarding the articles that did exist—OpenAI could not identify any facts or circumstances that would have prompted news organizations to look for those articles or put them on notice for potentially infringing content. The court likewise rejected OpenAI’s argument that The New York Times, as a “sophisticated publisher,” had a duty to take prompt action after being put on notice of the alleged infringement because (1) OpenAI could not establish that The New York Times was in fact on notice, and (2) the Second Circuit has squarely rejected a heightened “sophisticated rightsholder” theory of constructive knowledge.

The district court also declined to dismiss the contributory copyright infringement claims. The news organizations argued defendants materially contributed to and directly assisted with the direct infringement by building and training their LLMs with the news organizations’ works, deciding what content is outputted through specific training techniques, and developing LLMs capable of distributing copies of the news organizations’ works without their authorization. 

The district court concluded that the news organizations’ complaints sufficiently alleged defendants’ material contribution by citing “widely publicized” instances of copyright infringement after Open AI’s product ChatGPT and other software were released. Plaintiffs also included more than 100 pages of infringing output examples in their complaints. 

The district court similarly concluded the news organizations plausibly alleged defendants possessed constructive—if not actual—knowledge of end-user infringement. The complaints referenced reports that defendants’ LLMs were providing end users with copyrighted content and pointed out that the Times even notified defendants prior to filing suit that their tools were infringing copyrighted works. The district court also pointed out that defendants possessed actual or constructive knowledge of third-party infringement because the infringement was “central to [defendants’] business model” due to its significant reliance on copyrighted works.

Finally, the district court rejected the defense argument that the contributory infringement claims should still fail because defendants’ products are capable of “substantial noninfringing uses.” It determined that the cases defendants relied upon did not apply to the present dispute for a number of reasons, including that they (1) were decided on summary judgment or after trial rather than on motions to dismiss a complaint, (2) analyzed claims of contributory infringement by inducement rather than by material contribution, (3) did not present an “ongoing relationship” between the defendants and the end users, and (4) did not involve products that were built on the purportedly appropriated works.

Regarding the alleged DMCA violations, the news organizations argued the defendants intentionally removed or altered copyright management information (CMI) from their content and distributed content knowing the CMI was removed or altered without authorization. Defendants moved to dismiss both on standing grounds and on the merits. 

The district court first held the news organizations’ allegations of harm are sufficiently concrete to establish Article III standing. DMCA Section 1202(b) requires that a defendant know or have reasonable grounds to know their conduct “will induce, enable, facilitate, or conceal” copyright infringement, which “ensures that any violation of the DMCA is tied to concerns of downstream infringement.” The news organizations also have statutory standing to bring their DMCA claims. Plaintiffs have not merely alleged that defendants violated section 1202(b); they alleged that the violation injures them by concealing defendants’ own copyright infringement, enabling and facilitating end users’ copyright infringement, diverting users from their websites, and causing a decline in subscription and licensing revenue, the district court concluded.

As to the merits, defendants alleged the news organizations failed to allege any actual removal of CMI, that the defendants knew removing CMI would lead to infringement, or that defendants distributed copies of the news organizations’ works. The district court determined that all three complaints failed to state a DMCA claim against defendants. 

First, it found that the complaints failed to state a 1202(b)(1) claim against Microsoft. None of the allegations concerning Microsoft—including its partnership with OpenAI to develop certain tools and its provision of the cloud computing system OpenAI uses to train its models—related to any alleged removal by Microsoft of CMI from the news organizations’ works. 

Regarding the 1202(b)(3) claims against both Microsoft and OpenAI, the news organizations contended that the “regurgitations” generated by the LLMs constituted “distributions” of copies of their works. However, plaintiffs failed to show that the regurgitations were “works,” “copies of works” or “substantially similar or entirely reproduced.” The district court said the outputs were merely “excerpts” of the news organizations’ works and not copies for purposes of 1202(b)(3). 

In their unfair competition by misappropriation claims, the news organizations alleged defendants used their works without their authorization to train the LLMs, which in turn produced informative text of the same general type that plaintiffs produce. Accordingly, defendants’ directly compete with the news organization’s content and “free ride” on their efforts to gather time-sensitive news content. Defendants countered that the news organizations’ misappropriation claims are preempted by Section 301 of the Copyright Act. The district court sided with defendants, noting that Section 301 preempts most common law misappropriation claims involving copyrighted works and the news organizations failed to demonstrate that any exceptions applied.

The district court additionally denied OpenAI’s motion to dismiss the news organizations’ federal trademark dilution claims. Plaintiffs alleged that defendants used their distinctive and famous marks without authorization and used “lower-quality and inaccurate writing,” thereby diluting the quality of those trademarks. OpenAI countered that the complaints failed to allege the news organizations’ trademarks were “famous” under the Lanham Act. The district court concluded the news organizations sufficiently supported their allegations with detailed, factual descriptions of the nature and scope of the trademarks’ “widespread circulation, recognition, achievements, and consumer subscriptions,” which established that such marks are “famous” within the meaning of the statute. 

Summary prepared by David Grossman and Elena De Santis