Explosive AI Lawsuit: John Carreyrou Leads Authors Against Six Tech Giants in Copyright Battle
A group of authors led by John Carreyrou has filed a sweeping lawsuit against six AI leaders, arguing their models were trained on pirated books without permission or payment. The case goes straight at how training data is sourced and who profits from creative work.
For legal teams and writers, this is about leverage. Licensing, royalties, and the bounds of fair use are no longer theoretical-they're about to be tested in court.
What's Behind This Lawsuit
- Targets: Anthropic, Google, OpenAI, Meta, xAI, and Perplexity.
- Claim: Systematic copyright infringement via training on pirated digital copies of books.
- Why it's different: A coordinated action against multiple industry leaders at once.
- The legal pivot: It challenges a prior ruling that created a path for training on unlawfully obtained copies.
The Anthropic Settlement Controversy
The suit follows backlash to a proposed Anthropic settlement that would pay eligible writers about $3,000 each out of a $1.5 billion fund. Authors argue that payout ignores the commercial value extracted from their work and sets a weak precedent.
"LLM companies should not be able to so easily extinguish thousands upon thousands of high-value claims at bargain-basement rates, eliding what should be the true cost of their massive willful infringement."
Who's Being Sued-and What They're Selling
- Anthropic - Claude AI (valuation: $15B+)
- OpenAI - ChatGPT, GPT-4 (valuation: $80B+)
- Google - Gemini, Bard (valuation: $1.7T+)
- Meta - Llama models (valuation: $800B+)
- xAI - Elon Musk's AI company
- Perplexity - AI-powered search
Why AI Training Practices Are Under Scrutiny
According to the filing, the defendants used pirated books to train revenue-generating systems, without licensing or compensation to the authors. That raises three hard questions professionals will care about:
- Do AI companies need explicit licenses for training on copyrighted works?
- How should royalties be structured for training uses that aren't traditional reproduction?
- Where is the line between fair use and commercial exploitation?
The Legal Precedent Problem
In a prior case, a court distinguished between the initial piracy (illegal) and model training using those copies (treated as permissible once obtained). Authors say that logic ignores the economic use of their books to build valuable products. As they put it, the earlier settlement "seems to serve [the AI companies], not creators."
Why This Case Matters for Legal Teams and Writers
- Licensing: Expect pressure for data licensing agreements specific to training sets.
- Royalties: New models for training-use royalties could emerge, distinct from reproduction or adaptation.
- Fair use: Courts may clarify how fair use applies to model training at commercial scale.
- Industry norms: Publishers, agencies, and platforms may adopt standard AI clauses.
Action Steps for Authors
- Register your copyrights; keep clean records of editions and publication dates.
- Scan major pirate sites and data dumps for your works; preserve screenshots and hashes.
- Track AI outputs that quote or summarize your text verbatim; save prompts and results.
- Discuss contract language with your agent: AI training licenses, audit rights, and retroactive use.
- Consider class participation vs. opting out if you believe individual claims are stronger.
Action Steps for In-House Counsel and Publishers
- Run provenance audits on any internal or vendor-provided datasets; memorialize the chain of custody.
- Update contracts: define "training," "fine-tuning," and "embedding"; set license scope and revocation rights.
- Add indemnities for third-party data sources; require logging and deletion on breach.
- Create model-use policies: quote length limits, attribution, and disallowed prompts.
- Prepare a claims playbook: DMCA notices, settlement frameworks, and damages models.
FAQs: Quick Answers
Who is John Carreyrou and why is he involved?
He is a Pulitzer Prize-winning journalist who exposed the Theranos scandal in "Bad Blood." His participation brings visibility and credibility to the case.
Which companies are being sued?
Anthropic (Claude), Google (Gemini/Bard), OpenAI (GPT models), Meta (Llama), xAI, and Perplexity.
What was the previous Anthropic settlement about?
A proposed class settlement offering roughly $3,000 per eligible author from a $1.5B fund. Many authors call it inadequate, prompting this broader action.
How might this affect future AI development?
Companies may need explicit licenses for training data. Costs could rise, but it could create steady revenue streams for authors and publishers via licensing and royalties.
What to Watch Next
- Motions to dismiss: how courts treat training on unlawful copies.
- Class certification: whether claims proceed as a group.
- Discovery: disclosures on dataset sources and provenance controls.
- Regulatory signals: the U.S. Copyright Office's work on AI and fair use.
For ongoing policy context, see the U.S. Copyright Office's AI initiative here. The Authors Guild's AI advocacy hub is here.
Level Up Your AI Fluency (For Better Negotiations)
If you negotiate licenses, draft clauses, or advise authors, you'll need a working grasp of how training pipelines and model use actually work. Explore role-based programs here.
Conclusion
This case doesn't just ask who gets paid-it asks what counts as fair use when AI systems are built on books. The decision could set the rules for licensing, royalties, and accountability across the industry. Authors want payment and control; AI firms want clear, workable guidelines.
Watch the filings. Update your contracts. Document everything. The terms set here will influence how creative work is valued for years to come.
Your membership also unlocks:
