- Elsevier vs Meta: first science publisher sues over scraped research papers (Nature) — Elsevier joined a class-action lawsuit against Meta (filed May 5, 2026, SDNY) alleging use of millions of academic papers, books, and written works to train the Llama model. Co-plaintiffs: Cengage, Hachette, Macmillan, McGraw Hill, and author Scott Turow. The science publisher entry is significant: previous suits focused on news publishers (NYT) and fiction authors. Academic/scientific content raises distinct issues — much of it was publicly funded research.
- Major Publishers File Copyright Lawsuit Against Meta Over AI Training Practices (Influencer Magazine) — Additional context on the publisher group: this is framed explicitly as a coordination move — publishers comparing notes on the LibGen dataset Meta allegedly used for Llama training. The dataset contains pirated copies of millions of books, which is why Meta faces both copyright infringement and digital piracy claims simultaneously.
- Beyond the Training Data: The Shifting Battleground in AI Copyright Law (Bochner PLLC) — The litigation front is shifting: the original “training data = infringement” argument is being supplemented by output-side claims (AI-generated content that reproduces protected expression) and tool-side claims (AI systems designed to produce infringing outputs). Three distinct legal battlegrounds now, not one.
Case Tracker & Precedents #
- AI Lawsuit Tracker (2026) (AI Lawsuit Tracker) — Community-maintained tracker: 164+ active AI copyright litigation cases as of May 2026. Useful reference for tracking case status across the publisher, news, image, and code-training dimensions.
- Bloomberg Copyright Lawsuit Over AI Training Data to Move Forward (DiCello Levitt) — Bloomberg’s suit cleared a preliminary hurdle and proceeds to discovery. Bloomberg’s position is distinct from the news publisher suits: they are arguing that financial data (terminal data, news articles) is a specific category of proprietary commercial content that AI companies have systematically extracted without payment.
- AI in litigation series: An update on AI copyright cases in 2026 (Norton Rose Fulbright) — Thomson Reuters v. Ross Intelligence: summary judgment for Thomson Reuters at trial; Ross Intelligence’s fair-use defence failed; Third Circuit appeal now in progress. If the Third Circuit upholds, it will be the first binding appellate precedent that using protected content to train AI is not fair use. Timeline: decision expected Q3 2026.
Cross-links #
- Claude Integrations Thomson Reuters is simultaneously winning a copyright suit against AI training (Ross Intelligence) and partnering with Anthropic to build AI legal tools (CoCounsel). The distinction they’re drawing: training on copyrighted content without permission vs. licensed runtime access via MCP. The Third Circuit will test whether that distinction holds.
- Emerging theme: The litigation front is widening from books/news → academic/scientific publishing → financial data. Each content type brings a distinct set of plaintiffs, licensing norms, and legal arguments. Worth tracking whether the academic content suits are treated differently given publicly-funded research origin.
- Keyword suggestion:
"LibGen" meta llama training — the pirated dataset angle in the Meta suits is distinct from the fair-use argument and likely to generate specific legal findings.