IP Law Daily, COPYRIGHT NEWS: Microsoft, OpenAI strike back at NY Times’ infringement complaint, (Mar 8, 2024)

By Thomas Long, J.D.

In motions to dismiss, AI tech innovators argue that training processes for large language models don’t support contributory copyright infringement claims or DMCA claims for removal of copyright management information.

Pioneering developer of “large language model” (LLM) artificial intelligence technology OpenAI, Inc., and its primary investor, Microsoft Corp., have fired back at copyright infringement allegations made by The New York Times Co. in the newspaper company’s complaint filed in December 2023 in the federal district court in New York City (New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195). OpenAI and Microsoft have filed separate motions to dismiss, both of which target counts in the complaint that assert contributory infringement, violation of the Digital Millennium Copyright Act (DMCA), and state-law misappropriation of property. According to both defendants, the complaint fails to plausibly assert direct infringement upon which claims for contributory infringement can be based. Additionally, they argue, it fails to set forth DMCA claims for removal of copyright management information, both by failing to allege that such removal occurred and that the defendants had the requisite mental state for liability. They urge the court to dismiss the state-law claims because they are preempted by the Copyright Act.

OpenAI indirectly owns and controls ChatGPT—possibly the most popular and well-known LLM product at this time. Microsoft has invested at least $13 billion in OpenAI and owns a stake in the company. Microsoft also operates AI-powered tools, such as the Bing search engine and Bing Chat, recently rebranded as “Copilot,” which is an LLM similar to ChatGPT. The Times’ complaint accuses OpenAI and Microsoft of unlawfully using Times-owned copyrighted news articles and other works in “training” datasets to power their LLM products. The Times’ complaint alleges that the LLMs are capable of reproducing that content verbatim in response to users’ prompts, which amounts to direct infringement by the users, for which the defendants are contributorily liable.

Motions to dismiss. OpenAI filed its motion to dismiss with supporting memorandum of law on February 26. Microsoft filed its partial motion to dismiss the complaint on March 4, along with a supporting memorandum. The memoranda make similar legal arguments, although the Microsoft filing contains more details. Microsoft’s arguments depend in part on an analogy between LLM technology and the VCR (video cassette recorder, along with other innovations that were considered “disruptive” when they were first popularized, player pianos, photocopy machines, and personal computers), asserting that copyright poses no “obstacle” to the technology’s development and adoption. In the introduction of its memorandum, it invokes the language of a fair use defense, asserting that data used to train LLMs “does not supplant the market for the works.” Microsoft also asserts that the use made of the data is transformative—noting that the software is called a “transformer.” (“ChatGPT” stands for “Chat Generative Pre-trained Transformer”).

However, Microsoft’s motion doesn’t address the counts of the complaint targeting GPT-based models, but instead the counts based “on unsubstantiated suggestions that the public’s use of GPT-based products harms The Times.” According to Microsoft, the complaint does not reflect how LLMs are used in the real world. “The Times crafted unrealistic prompts to try to coax the GPT-based tools to output snippets of text matching The Times’s content—a technique The Times buries in lengthy exhibits to the Complaint,” Microsoft alleges.

Contributory infringement based on user outputs. Both OpenAI’s and Microsoft’s motions attack the complaint’s contention that the defendants are contributorily liable for direct infringement that occurs when users generate material with the AI tool that is substantially similar to Times-owned content. Contributory infringement claims depend on underlying acts of direct infringement. “But [the complaint] alleges no end-user infringement at all,” Microsoft argues, “and thus no knowledge on Microsoft’s part that end-users are making infringing use of LLMs that have a universe of entirely lawful uses.” In this respect, according to Microsoft, LLMs resemble the VCR, the substantial noninfringing capabilities of which were key in defeating copyright-based claims against marketing of that product.

Microsoft characterizes allegations in the complaint as “hypotheses” that someone could enter prompts into the GPT products that would yield output that is similar to The Times’ works. According to Microsoft, the only examples cited by The Times were deliberately conjured up by its counsel, and the complaint doesn’t point to any real-world acts of direct infringement. “The bare theoretical possibility that someone somewhere might engage in the same acrobatics The Times did here is not enough to plausibly allege direct infringement,” Microsoft asserts.

In addition, both OpenAI and Microsoft assert that the complaint fails to allege facts showing that they had the requisite knowledge of direct infringement. In Microsoft’s view, “the Complaint does not even assert that Microsoft has the ability to know that some user has engaged, is engaging, or will engage in an infringing use.” The fact that GPT products could be used to infringe is not enough, given the substantial noninfringing uses for those products, Microsoft repeats. OpenAI’s memorandum also points out that its terms of service prohibit uses of its service for copyright-infringing purposes, bolstering OpenAI’s argument that it would have had no reason to investigate direct infringement by users. Further, “even if OpenAI had investigated, nothing in the Complaint explains how it might evaluate whether these outputs were acts of copyright infringement or whether their creation was authorized by the copyright holder (as they were here),” its memorandum states.

DMCA Section 1202. The motions to dismiss next turn to The Times’ assertion that the defendants violated Section 1202 of the DMCA by removing copyright management information (CMI) from Times-owned works. OpenAI argues that the claim fails because The Times did not specify the CMI at issue, and it contends that there are no plausible allegations that any CMI was in fact removed. Microsoft focused more on the statute’s scienter requirements. For a Section 1202 claim to succeed, Microsoft points out, the plaintiff has to show that the defendant knew that removing the CMI would make it likely for infringement to occur (or to be harder to detect). “The Times never explains how this would happen in the real world,” Microsoft argues. “Its only examples are derived from its own prompts for particular Times content, which would require the user to already know the genesis of that content. And in any event, the outputs the Complaint cites are not copies of works at all, but mere snippets that are not cognizable under § 1202(b).”

The Microsoft memorandum cites a Second Circuit case referring to Section 1202’s intent/knowledge requirements as one for “double-scienter.” First, the defendant must intentionally remove or alter the CMI at issue. Second, it must know or have reason to know that this will “induce, enable, facilitate, or conceal an infringement.” According to Microsoft, the complaint fails to sufficiently allege the second part, containing only conclusory, “threadbare” assertions, without facts to back them up.

Repeating its contention that the only arguably infringing outputs in the complaint were “manufactured” by The Times and its attorneys, Microsoft argues, “The Complaint does not plausibly allege actual user conduct that yields infringing outputs in the first place, so it cannot possibly raise the inference that Microsoft somehow intended, knew about, or should have known about such behavior.” Microsoft further argues that anyone who intended to repeat the results cited by The Times in its complaint would have to purposely input or request New York Times content, making the presence or absence of CMI irrelevant, because the user would already know the source of the content.

Moreover, according to Microsoft, because of the nature of the LLM training process, the allegations in the complaint do not support a claim that CMI was unlawfully removed. The memorandum points out that courts have consistently rejected Section 1202 CMI removal claims involving excerpts of works, such as claims involving the mere “framing” of photographs in a way that crops out the CMI. “In short, a CMI-removal claim lies only with respect to copies of works that are ‘substantially or entirely reproduced,’” Microsoft argues (citation omitted).

Misappropriation claims—Copyright Act preemption. Finally, both defendants argue that state-law misappropriation claims asserted by The Times must be dismissed because they are preempted by Section 301 of the Copyright Act. Although some courts recognize a narrow exception to the general preemption of state-law torts that mirror copyright claims—specifically, the so-called “hot news” exception—Microsoft contends that the exception doesn’t apply here, even if The Times is attempting to invoke it. The complaint, by addressing copyrighted material, is concerned with the subject matter of copyright, even if The Times’ news content contains some noncopyrightable aspects (i.e., events and facts). In addition, the misappropriation count doesn’t depend on “extra elements” outside of the subject of copyright, such as the time-sensitive value of factual information or news, free-riding by the defendants, and a threat to the very existence of The Times from the challenged conduct. According to Microsoft, The Times alleges “neither the quantity of copying nor the immediacy of distribution necessary to sustain a ‘hot’ news claim.”

Statute of limitations. OpenAI’s motion includes a defense not raised by Microsoft—the Copyright Act’s three-year statute of limitations. This defense addresses The Times’ claims against OpenAI for direct infringement. According to OpenAI, the direct infringement claims are aimed at OpenAI’s activities to scour the web for content that it used to train the GPT products. “Because all those activities occurred more than three years ago … any claims based on them are time-barred,” OpenAI asserts. “Those claims are ‘stale,’ and the court should dismiss them so the parties can focus discovery on activities within the limitations period.”

Attorneys: Alexander Frawley (Susman Godfrey LLP) for The New York Times Co. Jeffrey S. Jacobson (Faegre Drinker Biddle & Reath LLP) for Microsoft Corp. Allison Levine Stillman (Latham & Watkins LLP) for OpenAI, Inc., OpenAI LP, OpenAI GP, LLC, OpenAI LLC, OpenAI OpCo LLC, OpenAI Global LLC, OAI Corporation, LLC and OpenAI Holdings, LLC.

Companies: The New York Times Co.; Microsoft Corp.; OpenAI, Inc.; OpenAI LP; OpenAI GP, LLC; OpenAI LLC; OpenAI OpCo LLC; OpenAI Global LLC; OAI Corporation, LLC; OpenAI Holdings, LLC

MainStory: TopStory AINews Copyright TechnologyInternet GCNNews NewYorkNews