COPYRIGHT AND AI: WHO OWNS THE RIGHTS?

Generative AI has forced two unsettled questions to the center of copyright law: who owns what AI makes, and who owns what AI learned from. So far the answer for purely machine-made work is “no one,” while the training fight plays out in dozens of lawsuits worth billions.

Thank you for reading this post, don't forget to subscribe!

Who owns what AI makes?

In its January 2025 report, the US Copyright Office reaffirmed that human authorship is the bedrock of copyright — so a work generated entirely by AI cannot be registered, and no one owns an exclusive right to it. Crucially, the Office concluded that “prompts alone do not provide sufficient human control” to make a user the author of the output. This was tested by Jason Allen’s award-winning image Théâtre D’opéra Spatial, made with 600+ Midjourney prompts: registration was refused for the AI-generated elements, though his manual Photoshop edits could qualify. The federal courts agree on the core point — in Thaler v. Perlmutter, they confirmed an AI cannot be the legal author. Where a human meaningfully shapes a work, that person owns their contribution; the machine-made parts fall into a copyright gap.

“Prompts alone do not provide sufficient human control to make a user the author of AI output.”

Who owns what AI learned from?

The bigger money is in the input. Model makers trained on vast troves of copyrighted text and images, and rights-holders are suing. The New York Times sued OpenAI and Microsoft (filed December 2023), seeking billions and alleging its articles were copied to build competing products. Anthropic agreed to a roughly $1.5 billion settlement after a judge found that ingesting millions of pirated, full-text books for commercial training went well beyond fair use. Getty Images, record labels, and authors including a New York Times investigative reporter have brought parallel claims against Stability AI, Suno, xAI, Meta, Google, and others.

The fair-use battleground

The central legal question — is training on copyrighted work “fair use”? — remains unresolved. As of late 2025 there were at least 51 active copyright suits against AI companies, and the handful of fair-use rulings split (roughly two for the AI companies, one against), often turning on whether the training data was lawfully acquired. The US Copyright Office’s third report, addressing training specifically, is still pending. Until courts or Congress settle it, the rules are being written case by case.

What it means for creators and businesses

The practical takeaways are sharp. If your team generates a logo, image, or article purely from a prompt, you likely cannot claim exclusive copyright in it — a competitor may be free to reuse it. To secure rights, keep a human meaningfully in the creative loop and document that contribution. And on the input side, businesses that build on models trained from unlicensed data inherit legal uncertainty until the courts rule. The throughline of this series holds: AI accelerates creation, but ownership and accountability still attach to humans.

Methodology & sources

• Copyrightability of AI outputs; human-authorship requirement — US Copyright Office (2025)
• Théâtre D’opéra Spatial / Jason Allen registration refusal — Copyright Office Review Board
• NYT v. OpenAI — OpenAI statement; Anthropic $1.5B settlement & lawsuit count — litigation tracker (2025)