Copyright and First Amendment Limits on Generative AI News Intermediaries

1. Introduction

In recent years, the integration of artificial intelligence (AI) into the journalism and media industries has transformed how information is produced and consumed. The launch of effective, penetrative platforms such as ChatGPT has led to a proliferation of AI-generated content and volatility within the media ecosystem. Generative AI is increasingly replacing human-form journalism and altering the entire journalism community. Recent studies estimate that around 9% of news articles in U.S. newspapers are at least partially generated by AI, often without disclosure, and that this rate is only increasing [1]. Stephen Reese, Professor of Journalism and Media at the University of Texas at Austin, coined the trend “a crisis of the institutional press,” and, indeed, the survival of the institutional press is difficult to predict [2]. The introduction of AI into journalism complicates a landscape of factors that frame news reporting, including linguistic diversity, cultural sensitivity, and the local-national relationship. There are also accessibility concerns for various demographics to utilize AI technology, thereby complicating the ability to achieve equitable access and informational independence [3].

Most importantly, the legality of AI in the press remains vague and problematic. The issue of evaluating AI in the news industry context is difficult for two reasons. First, AI involvement complicates Copyright Law doctrine. For instance, current laws contain a human authorship requirement, implying that a human must be claimed as the creator of a work in cases of infringement, but AI cannot comfortably satisfy this requirement. Second, AI involvement also clashes with First Amendment doctrine, particularly regarding the distinction between a publisher and a platform. In this article, I argue that generative AI news systems expose a fundamental mismatch in U.S. law: they function as human-like publishers when they substitute for journalism, but as platforms when they facilitate speech. Treating them as exclusively one or the other entails further problems in the other legal domain. Instead, this article argues that generative AI news systems must be classified differently across Copyright and First Amendment law to resolve this tension.

2. Generative AI as a Functional News Intermediary

Generative AI in the news industry serves as an “intermediary” program by processing, scraping, and reasoning to generate independent content for users. For a news reporter, generative AI can integrate seamlessly into almost every step of the editorial process: (1) newsgathering; (2) audience engagement; (3) news production. Regarding sourcing, major publications like AP, Bloomberg, and Reuters have been using computers and some degree of automation to scrape for new indications of news all over the world [4]. Then, a suite of AI tools helps news reporters write, edit, and publish content. Numerous programs, ranging from Amazon Polly to Datawrapper, allow reporters to generate graphics and figures, while natural language processing-based services like Otter and Trint accelerate the transcription process [5]. On the engagement front, news organizations have used AI algorithms from platforms like CrowdTangle and Chartbeat to analyze audience engagement and track trending topics on social media. These algorithms, scrapers, and programs have traditionally intervened in the work of the journalist, but only in a supplemental sense. What makes generative AI unique in this space is that it can deftly produce original content, not just verify or build on it [6].

According to the University of North Carolina, the implications of AI news production are already present. Automated reporting already generates full articles on routine topics such as sports scores, financial reports, and weather updates [7]. The New York Times put out guidance in May 2024, claiming that its editors will use generative AI as a “tool in service of their mission to uncover the truth” [8]. Yet, the relationship between a journalist and AI is not unlike the process of developing sources, in that AI systems often fail to serve their missions properly. Reports find AI-generated stories taking facts out of context, promulgating misinformation, and amplifying existing biases. AI systems maintain the capacity to hallucinate and ultimately lack a coherent understanding of the world [9]. Most of all, they obfuscate the ethical responsibility in reporting by diffusing accountability across insentient programs [10]. In an industry whose core tenet is to prioritize the truthful act of gathering facts and figures, the trend of implementing AI into the publishing of news content is controversial, not just because of potential inaccuracies, but also because it complicates the assignment of responsibilities for those inaccuracies.

3. Copyright Doctrine Fails to Account for AI-Generated Content

Under current copyright doctrine, AI in journalism fails the human authorship clause. Burrow-Giles Lithographic Co. v. Sarony (1884) established the need for human “intellectual conception” in establishing intellectual property of any kind [11]. In this case, “intellectual conception” refers to the author’s ability to determine what content appears, how it is expressed, and in what form it appears. Justice Miller, who wrote the unanimous decision in Burrow-Giles, reasoned that photographs displayed artistic creativity and, as such, fell outside of pure mechanical reproductions. Photographs could be copyrighted because the photographer made creative choices, thereby satisfying the human authorship requirement. Later, Feist Publications v. Rural Telephone Service (1991) defined the limits of copyright protection, stating that, for instance, “purely factual information arranged alphabetically” did not qualify as something original, since something original must contain a modicum of creativity [12].

Artificial intelligence, as a technology, functions outside of these definitions. AI lacks “intellectual conception” because the user provides a prompt and the model outputs original content through probabilistic processes. For example, one might seek to write an informative legal article, but it is the AI system that controls the output’s sentence structure, phrasing, framing, and, ultimately, the work itself. This is further complicated by Naruto v. Slater (2018), in which the Ninth Circuit upheld that any non-human entity cannot be designated as an original author of a work [13]. Humans may prompt the AI, but the AI is ultimately responsible for the final output. Therefore, because no human determines the expressive elements of AI-generated works, such outputs cannot satisfy the authorship requirement under existing copyright doctrine. As a result, assigning a “defendant” becomes difficult because it complicates the attribution of liability by obscuring who exercised control over the expressive elements of the work.

Without an entity for clear, identifiable authorship, these systems disrupt traditional mechanisms of accountability that underpin trust in the press, as backed by the law. In conventional reporting, authors can be named, evaluated, and held accountable for the accuracy and integrity of their work. Moreover, misinformation, hallucinations, and generally poor fact-reporting are often not viewed as reflections of news and media generally but rather on the human author behind the piece. AI-generated content diffuses this responsibility across opaque systems, making accountability difficult to assign.

4. Generative AI Defies Traditional First Amendment Definitions

Beyond human authorship concerns for assigning culpability in copyright cases, AI news generation affects First Amendment considerations regarding publishers and platforms. Reno v. ACLU (1997) provided needed definitions to determine the nature of a platform [14]. In Reno, the Court recognized that the Internet is a distinct media platform deserving of its own First Amendment protections, and that online media is subject to the same rights and restrictions as print media. Reno ultimately influenced statutes such as Section 230 of the Communications Decency Act, which protects online platforms from being held legally responsible for content posted by users. Under Section 230, online platforms are treated as intermediaries rather than publishers. AI systems struggle to hold up against Section 230 purview, though. When viewing AI generation in news through the lens of a platform, there are major inconsistencies. On one hand, platforms are protected when they host content or recommend content, as in the case of Zeran v. America Online (1997) or Gonzalez v. Google LLC (2023) [15]. However, in Fair Housing Council of San Fernando Valley v. Roommates.com (2008), the Ninth Circuit ruled that immunity does not extend to online services that materially contribute to unlawful content creation [16]. In short, since AI doesn’t just host but generates and shapes content, it exceeds the traditional legal protections surrounding platforms.

If AI can’t be governed under the current definitions of a platform, might it be better considered a publisher? In Miami Herald Publishing Co. v. Tornillo (1974), the Court unanimously held that entities cannot pressure the press to publish certain opinions [17]. Under Tornillo, publishing entails “editorial control and judgment.” Yet in AI-generated journalism, no identifiable human consistently exercises such control, thus calling into question the classification of these AI systems as traditional publishers. The question then becomes what counts as protected editorial judgment. Hurley v. Irish-American Gay, Lesbian and Bisexual Group of Boston (1995) sheds light on this question [18]. The unanimous decision held that private individuals or groups organizing expressive events cannot be legally required to include speech or participants altering the message they wish to present. Hurley implies that editorial discretion includes the selection and arrangement of speech. On the flip side, AI “selection” is not human expressive intent. Therefore, AI “selection” does not clearly constitute protected editorial judgment under Tornillo or Hurley. The result is a doctrinal gap, for AI-generated journalism fits neither the publisher model, which depends on human editorial control, nor the platform model, which assumes itself not to be a producer or generator of content.

5. Case Studies: Grappling with Generative AI in Journalism Contexts

Recent litigation involving news organizations provides a concrete lens through which to evaluate how courts are beginning to grapple with generative AI in journalistic contexts. In 2023, The New York Times alleged that OpenAI and Microsoft copied millions of its articles to train their large language models [19]. The Times argued that this form of scraping substituted for the newspaper’s work and harmed its subscription business. As the suit remains ongoing, the critical issue is the concept of “fair use” and whether large-scale text scraping for model training qualifies as “fair use.” The Times believes that regurgitating its journalism is not a transformative exercise, but OpenAI maintains that the training process leads to new insights rather than direct replication. Ultimately, the case is important because liability is the major concern for lawyers right now.

In addition to The New York Times, the Chicago Tribune has also mounted lawsuits against Perplexity AI [20]. These particular lawsuits considered the problem of regurgitating equivalent content. Perplexity argued that the act of copying was performed by the user, not the AI. The argument relies on the 2008 decision in Cartoon Network v. CSC Holdings, where the Court held that when technology automatically performs a function in response to a user's command, the user engages in direct infringement [21]. But, in this case, there were many possibilities for the alleged infringing act, from the initial copying of paywalled articles to the storage of those works in a retrieval system. Unlike in Cartoon Network, these steps are not user-designated but rather the coded output of the AI system.

6. Framework: Legal Accountability Regarding AI-Mediated Press

Generative AI in journalism highlights a failure in existing legal doctrines. On one hand, Copyright law presumes human authors exercising “intellectual conception,” while First Amendment doctrines clearly define publishers and platforms — neither of which seems to fit the use of the AI right now. Generative AI produces expressive content without a human author, while simultaneously operating as a creator and a platform. What this means is that authorship, liability, and accountability are all unclear in the context of the press.

Resolving these issues relies on developing a functional framework to classify the use of AI in the press. Two considerations in particular can help evaluate AI systems in legal contexts: (1) the degree of human control over the output, and (2) the degree of system contribution to the creation of the output. For the first consideration, courts should assess whether a human user meaningfully determined the expression of the output. Simple prompting should not warrant an attribution, but when such control is apparent and more involved, AI switches to a tool function, and traditional copyright protections should be sufficient to use as a baseline. Where human involvement is absent, though, the output falls outside what current copyright law can support. Under the second consideration, courts should evaluate the extent to which the AI system itself generates or shapes the content. The line should be drawn at the material contribution to the content, rather than merely hosting or transmitting it.

Together, these principles produce a more coherent allocation of legal responsibility. Where human control is high, and AI system contribution is low, liability and authorship should be attributed to the human. Where human control is low and AI system contribution is high, existing doctrines fail, and new forms of legal accountability must be considered. In intermediate cases, courts may need to adopt a hybrid approach.

7. Conclusion

Legal accountability must track meaningful control over expression. This is especially integral in the context of journalism, a discipline that amalgamates a desire to report truthfully through the dignity that is afforded a human newsgatherer. Without legal adjustment, generative AI will continue to erode the mechanisms through which the press is governed, leaving a growing domain of influential speech without any accountability framework. Thus, it is essential to have a functional framework to classify the use of AI in the press, providing nuance on when AI systems act as publishers versus platforms.

References

[1] Russell et al. 2025. "AI use in American newspapers is widespread, uneven, and rarely disclosed", https://arxiv.org/abs/2510.18774.

[2] Stephen Reese, The Crisis of the Institutional Press, Journalism Studies (2023).

[3] OECD, Artificial Intelligence, Digital Transformation, and Society (2023).

[4] Associated Press; Bloomberg L.P.; Reuters, reports on automation in news production (2023–2024).

[5] Amazon, Amazon Polly Documentation; Datawrapper; Otter.ai; Trint.

[6] Chartbeat; Meta Platforms (CrowdTangle), audience analytics tools documentation.

[7] University of North Carolina, Automated Journalism and News Production Report (2023).

[8] The New York Times, Guidance on the Use of Generative AI in the Newsroom (May 2024).

[9] Stanford University, AI Index Report 2024.

[10] Columbia Journalism Review, AI and Accountability in Journalism (2023).

[11] Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884).

[12] Feist Publications v. Rural Telephone Service, 499 U.S. 340 (1991).

[13] Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018).

[14] Reno v. ACLU, 521 U.S. 844 (1997).

[15] Zeran v. America Online, 129 F.3d 327 (4th Cir. 1997); Gonzalez v. Google LLC, 598 U.S. (2023).

[16] Fair Housing Council v. Roommates.com, 521 F.3d 1157 (9th Cir. 2008).

[17] Miami Herald Publishing Co. v. Tornillo, 418 U.S. 241 (1974).

[18] Hurley v. Irish-American Gay, Lesbian and Bisexual Group of Boston, 515 U.S. 557 (1995).

[19] The New York Times Co. v. OpenAI, Complaint (S.D.N.Y. 2023).

[20] Chicago Tribune v. Perplexity AI, ongoing litigation (2024).

[21] Cartoon Network v. CSC Holdings, 536 F.3d 121 (2d Cir. 2008).