Is Perplexity’s ‘no volition’ copyright strategy a clever legal shield or a long shot? Perhaps both
This article first appeared in Reuters Legal News and WestLaw Today.
When the New York Times and Chicago Tribune sued Perplexity AI in December 2025, the AI search startup faced a problem common to defendants in AI infringement actions. The suits claim Perplexity’s AI scrapes paywalled journalism and regurgitates it, sometimes verbatim, to its own paying customers.
Those allegations are not flattering. Most AI defendants have leaned heavily on fair use. Perplexity’s lawyers chose a different potential shield. They argue in their brief supporting a motion to dismiss that Perplexity did not “make” the allegedly infringing copies in a legally relevant sense and therefore engaged in no “volitional conduct” sufficient to support a claim of direct infringement.
In its motion to dismiss, Perplexity argued that it cannot be liable for direct copyright infringement because it never made a volitional choice to copy anything. According to the company, that act is performed by the user. The argument relies on the 2nd U.S. Circuit Court of Appeals’ 2008 decision in Cartoon Network v. CSC Holdings. There, the court held that when technology automatically performs a function in response to a user’s command, it is the user who engages in the volitional conduct required for direct infringement.
In simplified terms, the theory is that a user enters a query, the system automatically executes the request, but nobody at Perplexity consciously decides to copy the Times’ article in that moment. Courts have not previously evaluated this theory in the context of a generative AI search engine.
Even if successful, the argument would eliminate only direct infringement, leaving secondary liability in play. That distinction is not trivial as direct infringement is a strict liability standard. A plaintiff does not need to prove intent, knowledge, or even awareness. If the defendant made the copy, liability follows.
For a system designed to ingest and generate text drawn from across the web, that exposure is substantial. Each output that reproduces protected expression could constitute a separate act of infringement, potentially carrying statutory damages of up to $150,000 per work.
Secondary liability works differently. Contributory infringement requires knowledge and material contribution. Vicarious liability requires the right and ability to control the infringing conduct plus a direct financial benefit. Both doctrines introduce state of mind and control elements that are often fact intensive and harder to establish early in a case. Eliminating direct liability would therefore narrow the case meaningfully, even if it would not end it. Perplexity is also attacking the secondary infringement claims.
The obvious question on direct liability is whether the strategy is likely to succeed. The answer is probably not, at least not at the motion to dismiss stage. Beyond that point, the ultimate viability of the theory remains uncertain.
The Cartoon Network decision turned on a specific conception of causation. The court asked who “made” the copy in a proximate sense when a discrete work was reproduced. In that case the answer was straightforward. A particular subscriber pressed record, and a unique copy of a specific program was created for that subscriber. The service provider supplied the infrastructure, but the user supplied the initiating act that caused the work to be fixed.
Perplexity’s system presents a more complicated causal chain. What exactly is the allegedly infringing act? Possibilities include the initial crawling and copying of paywalled articles, the indexing and storage of those works in a retrieval system, the instantiation of text in RAM when a query is executed, or the generation of output that tracks the original expression. Each step reflects deliberate system design. Unlike the DVR in Cartoon Network, these steps are not user specific recordings of identified works. They are the predictable output of a system engineered to ingest, index, and reproduce textual content at scale.
If a court defines the infringing act as the reproduction and delivery of protected expression in response to a query, the argument that “the user made the copy” becomes less intuitive. The user asked a question. The system selected and generated expressive content drawn from copyrighted works. At that point the inquiry is not simply who pressed a button but who designed and built the machinery that predictably produced the allegedly infringing expression.
Courts applying the volitional conduct doctrine after Cartoon Network have often examined whether the service provider exercised control or made design choices that directed the copying. The more deliberate the system architecture, the harder it is to rely on user automation as a shield. A company that builds a machine capable of extracting and reproducing paywalled journalism has made many volitional choices, even if those decisions occurred at the design stage rather than at the moment of each query.
Perplexity will resist that characterization. It will argue that it built a general purpose answer engine rather than a system targeting specific publishers. It ingests large portions of the web, does not select works at query time based on editorial judgment, and does not intend to reproduce verbatim expression. Under that framing, any close similarity between output and a paywalled article is an incidental byproduct of model behavior triggered by a user’s prompt.
The difficulty with that position is that the volition doctrine has never required intent to infringe. The doctrine asks whether the defendant’s own conduct caused the copying. If a system is designed to crawl restricted content, retain it in usable form, and generate responses that substitute for the original article, a court could reasonably conclude that the provider’s conduct goes beyond passive facilitation. Even if no employee selects a particular article at the moment of a query, the system’s architecture may embody the relevant volitional act.
The newspapers’ attorney, Steven Lieberman, emphasized this point publicly. In a statement on March 1, as reported in Reuters, he noted that “Perplexity does not dispute [copying] The Times’s journalism from behind a paywall to deliver responses to their customers in real time.” See, e.g., “Perplexity asks court to trim New York Times copyright case,” Reuters, March 2, 2026. That framing highlights deliberate system design, paywall circumvention, and commercial delivery. Those elements are precisely the sort of upstream volition that could distinguish this case from Cartoon Network.
Another open question is whether the 2nd Circuit would extend Cartoon Network to a product that does more than store or transmit content. Generative systems actively synthesize and reproduce expression. The case for extending the earlier doctrine to that context appears weaker than Perplexity’s brief suggests.
Still, Perplexity may not need to win the volition argument outright in order to claim partial success. It may only need to make the theory plausible enough to complicate the plaintiffs’ path to summary judgment. If the court dismisses the complaint with leave to amend, the newspapers would need to plead specific volitional acts with greater precision. That process could narrow the set of works and conduct at issue.
Throughout its briefing, Perplexity also frames itself as defending information access against publishers attempting to monopolize facts. That rhetoric may have limited influence on a federal judge, but as public positioning, it reinforces the company’s effort to appear more like a search engine than a content pirate.
Ultimately the litigation may hinge less on volition and more on fair use. That doctrine assumes copying and asks whether it is legally excused. In many AI training cases defendants argue that training is transformative and non-substitutive because the output does not replace the original works. This dispute presents a more difficult version of that claim. When an answer engine provides a detailed summary or near verbatim account of a paywalled article in response to a user request for the same information, the substitution theory becomes concrete rather than theoretical.
The fourth fair use factor, market harm, is therefore likely to loom large. If the product enables users to obtain the substance of a paywalled article without visiting the publisher’s site or paying for access, a court may view the system as competing directly in the market for news consumption. Even sophisticated technology does not automatically justify reproducing expressive content for commercial gain in a way that plausibly substitutes for the original work.
That does not mean the fair use defense will fail, but it is fact intensive and unlikely to be resolved at the motion to dismiss stage. Over time, however, it may prove more consequential than the threshold dispute about volitional conduct.
The stakes of the volition argument extend beyond this litigation. If courts hold that upstream system design can satisfy the volitional conduct requirement, platform defendants may find it harder to rely on automation as a shield in disputes involving generative systems. If courts instead accept that user prompts break the causal chain for direct liability, plaintiffs in AI cases may be pushed toward secondary liability theories focused on knowledge, control, and financial benefit. Either outcome would significantly shape how copyright law adapts to generative technology.
Perplexity’s volition argument represents creative lawyering aimed at a genuinely novel legal question. Novel questions, however, carry risk for both sides. The same legal uncertainty that gives Perplexity hope may also make courts reluctant to grant a sweeping victory at the pleading stage.
The cases are Chicago Tribune Co. LLC v. Perplexity AI Inc. (1:25-cv-10094) and The New York Times Co. v. Perplexity AI Inc. (1:25-cv-10106) in the U.S. District Court for the Southern District of New York.
