Who Owns AI-Generated Content?

💰 Bartz v. Anthropic: $1.5B Settlement

The largest AI copyright settlement in history. Authors secured approximately $3,000 per book for 7 million copyrighted works used in training data without permission.

Key Takeaway: The settlement does NOT grant Anthropic a future training license. This sets a precedent for compensating rightsholders but leaves the fair use question unresolved.

🏛️ Supreme Court Declines Thaler Appeal

The Supreme Court declined to hear Thaler v. Perlmutter, cementing the principle that pure AI-generated works cannot be copyrighted under U.S. law.

Key Takeaway: Human authorship remains required for copyright protection. However, human curation and iterative editing of AI outputs CAN be copyrighted.

📋 Copyright Office 2025 Reports

Part 2 (Jan 2025): Reaffirmed that human authorship is the "bedrock" of copyright. First AI-assisted image registered in Feb 2025 required 35 iterative edits.

Part 3 (May 2025): Fair use for AI training is determined case-by-case. No blanket ruling either way.

Key Takeaway: Substantial human editing strengthens copyright claims. Document your creative process.

⚖️ 70+ Active AI Copyright Lawsuits

NYT v. OpenAI: Proceeding to trial. Court ordered disclosure of 20M ChatGPT logs (Jan 2026).

Thomson Reuters v. Ross: First judicial ruling AGAINST AI training as fair use.

Meta Llama case: Proceeding on allegations of mass copyright piracy in training data.

Key Takeaway: The legal landscape remains unsettled. Expect significant precedent-setting decisions in 2026-2027.

📝 Platform Terms Updates

Anthropic: Consumer terms now allow training by default (Aug 2025). Commercial API users retain full ownership + indemnification.

OpenAI: Settlement negotiations ongoing with multiple publisher groups.

Midjourney: Sued by Disney and Universal (June 2025) over character training data.

Key Takeaway: Read platform terms carefully. API/Enterprise tiers offer stronger protections than consumer plans.

🎵 Music Industry Escalation

BMG v. Anthropic (March 2026): $70M lawsuit claiming Claude was trained on pirated music lyrics and sheet music.

UMG/Concord v. Anthropic (Jan 2026): $3.1B lawsuit — the largest music copyright claim against AI to date.

Key Takeaway: Music copyright enforcement is even more aggressive than text/image. AI music generation faces heightened legal scrutiny.

⚖️ Fair Use Judicial Split

Out of the first 3 federal judges to rule on AI training fair use:

• 2 ruled FOR AI companies: Training is transformative use (not direct copying)
• 1 ruled AGAINST: Thomson Reuters case — training on legally licensed content ≠ fair use for competing product

Key Takeaway: Fair use determination hinges on whether training data was legally acquired and whether the AI competes with the original copyrighted work.

🎯 What This Means for Your Business

1. Pure AI outputs lack copyright protection. Add substantial human editing to strengthen your IP position.
2. Platform ownership terms vary dramatically. API/Enterprise plans offer stronger protections than consumer tiers.
3. Training data lawsuits are still unfolding. The fair use question won't be definitively settled until 2027+.
4. Music and video AI face higher legal risk than text-based AI due to aggressive rightsholder enforcement.
5. Document your creative process. If you're challenged on copyright, proof of human authorship is critical.
6. Consider IP indemnification. Platforms like Microsoft Copilot and Anthropic Enterprise offer copyright protection that consumer plans don't.