Corporate Counsel Daily, Dow Jones, New York Post sue Perplexity AI for ‘massive’ illegal copying and trademark harm from ‘hallucinations’, (Oct 23, 2024)

By Robert Margolis, J.D.

Complaint alleges copyright violations at the “input” and “output” stages, as Perplexity copies the publications’ content to include in its database, then provides users copies or paraphrases of copyrighted articles.

Alleging that Perplexity AI, Inc. has chosen to develop its generative AI technology by “stealing copyrighted material” rather than by legally “recognizing the legitimate rights of copyright holders” through negotiated licensing agreements, the publishers of the Wall Street Journal and New York Post have filed a three-count complaint in federal district court in New York, alleging copyright infringement in violation of the Copyright Act and false designation of origin/trademark dilution in violation of the Lanham Act. The publishers contend that Perplexity “engages in a massive amount of illegal copying” of their copyright works, both in the “input” stage when Perplexity copies their works to populate the Perplexity database, and the “output” stage when Perplexity provides answers to users’ queries that include the publishers’ copyrighted material. The publishers also allege that Perplexity’s “outputs” combine copyrighted with “hallucinations,” AI generated content not from the publishers, all under citations using the publishers’ trademarks, which has the effect of falsely designating that the content came from the publishers and diluting their trademarks. The case was filed on Monday and has been assigned to Hon. Katherine Polk Failla (Dow Jones & Company, Inc. v. Perplexity AI, Inc., No. 1:24-cv-07984-KPF (S.D.N.Y. filed Oct. 21, 2024)).

Perplexity’s generative AI. The publishers allege “on information and belief” that to build a product that serves as a “substitute” for the “vast market for content” produced by professional journalists and which requires substantial effort, skill, experience, risk, and cost to create, and for which the publishers obtain revenue through advertising, paid subscriptions and licensing of content (including to other AI companies competing with Perplexity), Perplexity engages in “massive” copying of the publishers’ original, copyrighted content for use in its internal “retrieval-augmented generation (‘RAG’)” database. This includes, alleged again “on information and belief,” “all of” the publishers’ copyrighted content since Perplexity was launched in 2022.

Perplexity then allegedly uses the copyrighted material to generate responses to queries from users, which according to the publishers “are intended to and do act as a substitute for news and other information websites.” These responses, retrieved from the RAG database by use of an AI model (which the publishers call a “large language model”), are in the form of outputs that “summarize and paraphrase original, human-generated content, even at times reproducing that content verbatim.” (Emphasis in original.) The publishers note Perplexity boasts that its users can “Skip the Links” to the original publishers, and thus Perplexity purports to serve as a substitute for accessing the publishers’ copyrighted works on their own websites of from licensees. This diverts revenue from the publishers, they further allege.

The publishers have brought two counts of copyright infringement against Perplexity. The first count alleges infringement in the “input” stage by copying the publishers’ copyrighted content for inclusion in the RAG database. The second alleges infringement at the “output” stage, when responses to queries include copyrighted material. The complaint includes an example of an output Perplexity generated that provides the full text of an article from the New York Post. Notably, the publishers assert that the copyright violations alleged concerning the “input” stage are not dependent on whether Perplexity also violates copyright at the “output” stage. The initial copying without permission itself constitutes infringement, they allege, irrespective of whether a response to a user query also violates copyright.

Cited sources. The complaint includes a section addressing Perplexity’s responses to public criticism that it is ignoring copyright law. First, Perplexity has touted its “Cited Sources” feature, which includes citation to the original source, with a link. Notably, according to the publishers, Perplexity “actively encourages” its users to “skip” those links. Thus, the publishers allege that the citations to sources actually make it less likely that users will use the links to access original content, since the inclusion of citations makes the Perplexity-generated conduct seem more reliable. They note that despite Perplexity receiving approximately 250 million queries per month, the publishers have experienced virtually no click-through traffic on their websites from Perplexity’s cited sources links.

Fair use. The publishers also allege that at other times Perplexity’s principals have alluded to Perplexity’s use of content as falling within the fair use exception to infringement. The complaint includes preemptive strikes against such a defense, alleging: (1) Perplexity’s use is solely for a commercial purpose; (2) it copies “every single word” of the publishers’ original content “that it can get its hands on”; (3) its use of copies creates a commercial substitute for the publishers’ protected works; and (4) this harms the publishers’ traditional advertising and subscription revenue streams. Finally, they allege that there is nothing “transformative” about “massively copying” copyrighted works and “repackaging” it to consumers.

Hallucinations. The complaint also includes one count brought under the Lanham Act, alleging that Perplexity’s creation of “hallucinations” constitutes a false designation of origin and dilutes the publishers’ trademarks. The publishers allege that the Perplexity “outputs” not only include verbatim reproductions of copyrighted works or paraphrased summaries of copyrighted works, but also include “made-up text (hallucinations)” in its outputs and attributes those hallucinations to the publishers’ publications using the publishers’ trademarks. The publishers allege this has the effect of (1) confusing and deceiving users into believing the hallucinations are the work of, sponsored by, or approved by the publishers, when they are not, thus constituting false designation of origin; and (2) diluting the publishers’ marks by blurring and/or tarnishing, since the hallucinations were not created under the standards used by the publishers in creating their journalistic works. The complaint includes examples of Perplexity outputs that include both verbatim text from a New York Post article, as well as several paragraphs of “hallucinations,” which the publishers allege leads Perplexity users to believe that because the output as a whole is sourced to the New York Post, those “hallucinations” are the New York Post’s work.

Distinctions. The publishers also assert that what Perplexity does is distinguishable both from search engines generally and from other AI products operating with “legitimate” business models. Traditional search engines “merely provide links” to original sources of copyrighted works. Further, other AI companies have entered into licensing agreements with content providers. By not entering into such licensing agreements, “Perplexity has elected the illegal approach” to developing its AI technology, the publishers allege.

Remedies. For the copyright claims, the publishers seek injunctive relief against Perplexity further copying their works without authorization, or using databases, including its own, that include copyrighted works, to generate outputs. They seek destruction of any index or database that contains their copyrighted works, as well as destruction of any copies of works already made. They also seek statutory damages up to $150,000 for each infringement, actual damages, and disgorgement of Perplexity’s profits for each infringement. For the trademark claim, the publishers also seek an injunction against using their trademarks in a manner that dilutes, statutory damages, actual damages, and Perplexity’s profits. They also seek attorney fees.

The Case is No. 1:24-cv-07984-KPF.

Attorneys: William P. Barr and Paul T. Cappuccio (Torridon Law PLLC) for Dow Jones & Co., Inc.

Companies: Dow Jones & Company; Inc.; NYP Holdings, Inc.; Perplexity AI, Inc.

News: AINews Copyright TechnologyInternet Trademark GCNNews NewYorkNews