AI copyright fights escalate
Major copyright suits are piling up against AI firms—BMG is seeking up to $150,000 per song from Anthropic, while Britannica and Merriam‑Webster are suing OpenAI over alleged use of 100,000 articles—threatening how foundation models source training data. Parallel litigation could reshape dataset practices across embodied AI and robotics companies that rely on scraped human content. (tipranks.com) (technobezz.com)
BMG filed a copyright complaint against Anthropic in the U.S. District Court for the Northern District of California on March 17, 2026 (case no. 5:26‑cv‑02334). (law.com) The complaint lists roughly 493 allegedly infringed musical compositions and cites lyrics from artists including Bruno Mars, the Rolling Stones, Ariana Grande and Justin Bieber as exemplars. (ainvest.com) BMG alleges Anthropic harvested text through automated web scraping and “torrenting” pirate libraries, removed copyright‑identifying metadata during dataset cleaning, and trained Claude on that corpus so the model can reproduce lyrics on demand. (musicbusinessworldwide.com) Separately, Encyclopædia Britannica and Merriam‑Webster filed suit against OpenAI in the Southern District of New York on March 13, 2026 (case no. 1:26‑cv‑02097), alleging OpenAI ingested and reproduces Britannica and Merriam‑Webster content to train and augment ChatGPT. (courthousenews.com) Britannica’s complaint includes trademark claims under the Lanham Act and specifically accuses ChatGPT’s retrieval‑augmented generation of returning verbatim or near‑verbatim passages and falsely attributing content to the publishers. (techcrunch.com) These filings come after Anthropic’s September 2025 settlement with authors (reported at roughly $1.5 billion), a development legal advisers say has already pushed AI firms to strengthen data‑provenance records, licensing practices, and audit trails for training corpora. (ropesgray.com) Prominent law firms and industry memos are urging companies that build embodied‑AI and robotics stacks to document source licenses, run dataset audits, and prepare to produce training logs and model outputs in discovery—requirements repeatedly emphasized in post‑settlement legal guidance. (nortonrosefulbright.com) Plaintiffs in both cases are seeking injunctive relief, damages and discovery orders that would force disclosure of training datasets and model outputs, a procedural outcome lawyers say could compel robotics companies that rely on scraped human content to alter data‑collection pipelines and contracts. (msn.com)
