AI Copyright Law in the United States (2026 Guide)

Two questions dominate AI and copyright in the United States: can you own what an AI creates, and can you be liable for what it reproduces? Federal law answers each more clearly than the headlines suggest. This hub maps both, and links to detailed guides.

Information last verified on 2026-06-25. This article presents general legal information about United States federal law, not legal advice. The law on AI and copyright is evolving, and several cases cited below are still on appeal.

Jurisdiction scope: This hub covers United States federal copyright law as it applies to artificial intelligence: the ownership of AI-generated works, infringement liability, open-source licensing, and the training-data cases. It does not address patent or trade-secret law in depth, and it does not cover other countries. For software specifically, see the code guide linked below; for other countries, see how AI and copyright differ worldwide.

Can you copyright what an AI makes?

Generally not, unless a human supplied the creative expression. United States copyright protects only works of human authorship, a requirement the courts and the Copyright Office have applied consistently to AI output. In Thaler v. Perlmutter, the D.C. Circuit affirmed in March 2025 that the Copyright Act requires a human author, and the Supreme Court declined to review that ruling in 2026. The principle reaches every kind of work, not just art: AI-generated code, images, music, and text all face the same bar.

This does not mean AI-assisted work is unprotectable. The Copyright Office's January 2025 report concluded that a work is protectable to the extent a human exercised creative control over its expressive elements, and that prompting alone usually does not supply that control. When registering, applicants must disclose and disclaim purely AI-generated material under the Office's guidance at 88 Fed. Reg. 16,190. The takeaway for creators and companies is that ownership tracks human contribution, so the portions a machine generated on its own may belong to no one.

When does AI use become copyright infringement?

Infringement happens when protected expression is reproduced, distributed, or adapted without authorization, and AI does not change that analysis. Under 17 U.S.C. § 106, the copyright owner alone may reproduce a work, prepare derivatives, or distribute it; under § 501, anyone who violates those rights is an infringer. Critically, infringement is a strict-liability wrong: there is no knowledge or intent requirement. The Supreme Court confirmed as early as 1931, in Buck v. Jewell-LaSalle Realty Co., that "intention to infringe is not essential under the act."

For AI users, the risk is that a model reproduces training material in its output and the user then publishes it. The user is the one distributing the result, so the user carries the exposure, even if they did not know the output traced back to a protected source. Innocence is not a defense to liability; it only lowers statutory damages, which fall from a standard range of $750 to $30,000 per work to as little as $200 for an innocent infringer under § 504(c)(2), and rise to as much as $150,000 for willful infringement.

AI-generated code and open source

Software is where these rules bite first, because AI coding assistants are trained on vast amounts of open-source code and can reproduce recognizable pieces of it. Open source is not public domain; it is copyrighted code released under a license with conditions. Permissive licenses such as MIT and Apache 2.0 require attribution. Copyleft licenses such as GPL and AGPL require that derivative works stay open, and the AGPL extends that duty to software offered over a network. Courts enforce these terms through copyright (Jacobsen v. Katzer) and as contracts (Artifex v. Hancom), so shipping copyleft code inside a closed product without complying is both infringement and breach.

Because this is the most common real-world dispute, we cover it in depth in a dedicated guide: is AI-generated code copyright infringement?. It explains strict liability, the open-source license families, who is liable when a tool like Copilot suggests infringing code, and why purely AI-written code may not be copyrightable at all.

The training-data fight: is it fair use?

This is the most unsettled question, and the courts are split. Training a model requires copying enormous quantities of existing work, and whether that copying is fair use under 17 U.S.C. § 107 is being litigated case by case. In Thomson Reuters v. Ross Intelligence, a court rejected a fair-use defense where a non-generative AI was trained on a competitor's copyrighted material to build a directly competing product. In Bartz v. Anthropic and Kadrey v. Meta, courts found that training large language models on books was transformative and therefore fair use, although Anthropic still faced liability for sourcing pirated copies to begin with.

Several of these proceedings remain active, and Bartz v. Anthropic ended in a settlement, so none produced a final appellate ruling. Each turned heavily on its facts, especially on whether the AI's output competes with the original in the market. For now, the safest reading is that lawful acquisition of training data and genuinely transformative use help a fair-use defense, while building a market substitute from someone else's protected work undermines it.

Federal, not state: why there is no "AI copyright by state"

Copyright is one national system. Section 301 of Title 17 preempts any state-law right equivalent to copyright in a fixed work, so the same federal Copyright Act applies in every state. There is no California AI-copyright statute or Texas AI-copyright statute that changes who owns or infringes a work. What does vary by state are adjacent claims that often travel with a copyright dispute: trade-secret misappropriation under each state's Uniform Trade Secrets Act, and breach of a software license under state contract law. Those are separate from copyright, and they are the reason a single incident can produce claims in both federal and state court.

What is settled and what is still moving

Where to go next

For the detailed treatment of software, start with is AI-generated code copyright infringement?, the keystone guide in this hub. To compare how other countries handle AI and copyright, including Europe's stricter training-transparency duties, see how AI and copyright differ worldwide. For the separate question of licensing open-weight AI models inside a product, see AI model licensing and open-source attribution. For state-level AI regulation beyond copyright, such as disclosure and deepfake rules, see AI laws and regulation in the United States.

This is general legal information, not legal advice. It describes United States federal copyright law as of 2026-06-25 and does not address your specific facts. The law on AI-generated works is evolving, and several cited cases remain on appeal. Consult an attorney licensed in your jurisdiction before acting on AI-generated or open-source material.

Related articles

• Is AI-generated code copyright infringement?
• How AI and copyright differ worldwide
• AI model licensing and open-source attribution
• AI laws and regulation in the United States

Last updated: 2026-06-25. Statutes and cases cited reflect their status as of 2026-06-25; several AI-related decisions remain on appeal.