CNN recently filed a lawsuit against Perplexity AI, alleging that the company copied more than 17,000 CNN articles, videos, and images to power its AI search products without permission or compensation. The case arrives amid a broader wave of AI copyright litigation, and the instinct among policymakers is likely to treat it as more of the same. That instinct is wrong, and treating this case as equivalent to other training-data lawsuits could lead to poor policy in either direction.
Publisher copyright claims should not be used to broadly constrain AI development. However, this CNN case is legally distinct from the training-data lawsuits filed against OpenAI, Anthropic, and others, and that distinction matters. Those cases center on whether using publicly available content to train AI models constitutes fair use. CNN alleges something different. It claims that Perplexity’s products reproduce CNN’s journalism in near-verbatim form and in real time through retrieval-augmented generation (RAG), a technique that pulls live content from specific websites or databases to answer Internet search queries and reduce AI “hallucinations.” As a result, Perplexity may be delivering content that is identical or substantially similar to material CNN reserves for both visitors and paying subscribers, without directing users to CNN.com. The good news is that existing copyright law already addresses this issue. Congress does not need to wait for a new legal doctrine to emerge—the necessary legal tools are already in place.
CNN also makes serious trademark allegations, which may be the stronger part of its case. According to the complaint, Perplexity allegedly marketed its “Comet Plus” subscription tier in a way that allowed users to access CNN’s premium content, despite the absence of any commercial partnership between the two companies. Fair use can serve as a valid defense against copyright infringement, but not against commercial misrepresentation. The law is clear that a company cannot use another brand’s content in a manner that falsely implies endorsement. In addition, CNN also alleges willful infringement, citing Perplexity’s purported failure to comply with a formal cease-and-desist letter in December 2025. If true, this fact would be a critical element in a finding of willful infringement and would undermine Perplexity’s good-faith defense.
The decline of local news has significant implications for journalism, communities, and democratic governance. Some policymakers have proposed addressing this issue by requiring AI Companies to pay licensing fees for news content. However, the evidence does not support that approach. Since 2005, the U.S. news industry has lost more than 3,200 newspapers, a trend driven largely by the collapse of print advertising that predates AI by well over a decade. The rise of AI answer engines compounds an already existing problem. According to the content-licensing platform TollBit, in 2025, AI search tools generated 96 percent less referral traffic for publishers than traditional Google search. Yet the Journalism Competition and Preservation Act and similar mandatory-licensing proposals, unfortunately, do not target that problem effectively. Statutory licensing does not resolve the power imbalance it claims to address because pool distributions tend to flow disproportionately to the largest publishers rather than to the local outlets most at risk.
International experience offers a useful lesson. Canada’s Online News Act produced a predictable outcome: When governments mandate payment terms without providing platforms a workable path to compliance, platforms with exit options are more inclined to use them. The Media Ecosystem Observatory, a joint project of McGill University and the University of Toronto, estimated that Canadian news publishers saw approximately eleven million fewer daily views as a direct result, with local outlets bearing the greatest losses. Those are precisely the publishers the legislation was designed to help. The lesson from Canada is not that mandatory licensing is inherently wrong. Rather, it is that poorly designed mandatory licensing can fail and disproportionately harm the small publishers they are meant to protect. Legislation that mandates payment terms without offering platforms a workable compliance option creates incentives for larger players to restructure their services on their own terms. The United States should take a more careful approach.
Voluntary AI content licensing is emerging as a viable model, but it remains fragile and depends on solid legal infrastructure to ensure durability. Gannett, TIME, Le Monde, and other publishers have already reached licensing agreements with Perplexity, demonstrating that mutually beneficial commercial deals between AI firms and news organizations are possible. However, the CNN case illustrates that copyright litigation is not the alternative to a licensing market—rather, it is often the legal backstop that encourages companies to negotiate voluntary agreements in the first place.
The right policy response is targeted, not sweeping. The FTC already has the tools to act on the false-affiliation claims alleged in this case under Section 5 of the FTC Act. Since 2024, the agency has brought at least a dozen AI-related misrepresentation cases under its “Operation AI Comply” initiative—with no new legislation required—and it should continue to use those tools if these allegations are proven to be true. Although Congressional action does not appear necessary in this instance, if Congress were to act, it should simply clarify that near-verbatim AI reproduction of paywalled and otherwise copyrighted content for commercial purposes falls outside the scope of fair use.
Further, if this clarification were made, Congress should also ensure fair use latitude for model training, where courts are already developing a workable doctrine. To ensure that these recommendations are enforceable rather than aspirational, Congress should establish a safe harbor that would create legal certainty for developers who both respect machine-readable opt-out signals and use automated tools to filter sensitive personal information. These are narrow, targeted adjustments that address real harms without dismantling the open information ecosystem that underpins American AI leadership or preempting courts that are already doing their jobs.
The CNN-Perplexity case is a reminder for policymakers to enforce existing laws precisely, while building the legal and commercial infrastructure needed to make alternatives to litigation not only practical, but viable. The goal should be to create the conditions for AI answer engines and news publishers to operate as commercial partners rather than as adversaries. That requires a well-designed licensing framework that provides local newsrooms with a durable revenue stream from AI distribution, and where AI products compete on the quality of their sourcing. Achieving that outcome calls for both targeted enforcement now and affirmative policy design next.
Image credit: Canva
