Anthropic's $1.5B settlement approved as court distinguishes piracy from AI training
A federal judge provided final approval for a $1.5 billion settlement between Anthropic and various authors and publishers regarding the illegal collection of training data. The ruling maintains that training AI models on legally obtained text constitutes fair use, though it establishes that the use of pirated content for training remains subject to legal challenges.
Key Takeaways
- Authors and publishers will receive approximately $3,000 per work across an estimated 500,000 titles.
- Judge William Alsup ruled that training AI on legally obtained books is transformative and constitutes fair use.
- The 91% participation rate from eligible rightsholders helped overcome 54 formal objections to the payout size.
- Anthropic is required to destroy the specific datasets identified as pirated as part of the agreement.
- The settlement avoids a trial but creates no binding legal precedent for pending cases against OpenAI or Meta.
Why It Matters
The approval cements a critical two-track legal framework for the AI industry: high protection for the act of training combined with high liability for data sourcing. By isolating the 'piracy' issue from the 'fair use' issue, the court has signaled that AI developers cannot use transformative training as a retroactive shield for illegal data acquisition. For the streaming and media ecosystem, this reinforces the urgency of transparent data provenance. While this specific case is closed, the lack of an appellate ruling means the core fair use question remains unsettled in other jurisdictions. Watch for the consolidated OpenAI litigation in New York for the next potential challenge to this training-as-fair-use standard.
Additional Context
The Anthropic settlement arrives as publishers increasingly shift from litigation to high-value licensing agreements. In May 2024, News Corp signed a five-year deal with OpenAI worth an estimated $250 million, per the Wall Street Journal. This was followed by a March 2026 agreement between News Corp and Meta, reportedly valued at $50 million per year, according to everything-pr. These deals reflect a growing market where content owners prioritize recurring revenue over the uncertain outcomes of copyright trials, especially as judicial trends in the Northern District of California appear favorable toward the technical process of AI training. Related legal battles continue to test these boundaries. In July 2026, major publishers including Hachette and Cengage filed a class action against Google, alleging that the Gemini platform was trained on books originally provided to Google Books for search snippets only, per The Daily Star. Meanwhile, the Southern District of New York remains a critical venue; in January 2026, Judge Sidney Stein ordered OpenAI to produce 20 million ChatGPT user logs in the discovery phase of the consolidated author and publisher lawsuits, according to everything-pr. These logs are expected to reveal whether AI models generate infringing outputs, a key factor that Judge Alsup noted was absent in the Anthropic case. Beyond text-based models, the industry is closely monitoring the music and video sectors. Universal and Warner settled disputes with AI music firms Suno and Udio in late 2025 by converting infringement claims into licensing partnerships, per troveo.ai. Similarly, Disney recently signed a licensing deal with OpenAI for its Sora video generator while simultaneously pursuing a $150,000-per-work infringement claim against Midjourney over character imagery. This dual-track strategy—licensing to established partners while suing unauthorized developers—is becoming the standard operating procedure for major media rightsholders navigating generative AI.
Read full article at techcrunch.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source