Is AI-Generated Code Copyright Infringement? (2026)

Using an AI coding assistant does not shield you from copyright law. If the tool reproduces protected code and you ship it, you can be liable under 17 U.S.C. § 501, even without knowing. And purely AI-written code may not be copyrightable at all, so you may have nothing to protect.

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-generated works is evolving, and several cases cited below are still pending.

Jurisdiction scope: This article addresses United States federal copyright law (Title 17 of the U.S. Code), the human-authorship requirement, and open-source license compliance for software, including AI-assisted and AI-generated code. It does not cover patent or trade-secret law, and it does not address other countries. For the global picture, see how AI and copyright differ worldwide.

For the wider United States picture and how these rules fit together, see our AI, copyright, and code hub.

Is using AI to write code copyright infringement?

Using an AI assistant to write code is not, by itself, copyright infringement. The legal problem arises when the tool reproduces someone else's protected expression and you copy, distribute, or build on it. Under 17 U.S.C. § 106, only the copyright owner may reproduce a work, prepare derivative works, or distribute copies. Under 17 U.S.C. § 501, anyone who violates those rights is an infringer. The statute sets no knowledge requirement, so infringement does not turn on whether you realized the output came from protected code.

Two separate questions follow, and people routinely blur them. First, did the AI reproduce protected expression that you then shipped? If so, you may be liable regardless of intent. Second, can you own what the AI produced for you? Often you cannot, because United States law protects only human authorship. The sections below work through both questions, then turn to the open-source license rules that decide most real-world disputes.

Copyright is federal, not state: one national rule

There is no state copyright statute for software. Copyright is governed exclusively by federal law. Section 301 of Title 17 preempts any state-law right that is equivalent to copyright in a fixed work, so California, Texas, and New York all apply the same federal Copyright Act to code. This is why "copyright by state" guides do not exist: the rule does not vary from one state to the next.

"On and after January 1, 1978, all legal or equitable rights that are equivalent to any of the exclusive rights within the general scope of copyright as specified by section 106 ... are governed exclusively by this title." 17 U.S.C. § 301(a)

What can vary by state are the related claims a plaintiff adds alongside copyright. Trade-secret misappropriation runs under each state's version of the Uniform Trade Secrets Act, and breach of a license agreement runs under state contract law. Those are separate causes of action layered on top of the single federal copyright rule, not state copyright law.

"I did not know the AI copied it" is not a defense

Copyright infringement is a strict-liability wrong. Section 501 makes anyone who violates an exclusive right an infringer, with no element of intent or knowledge. The Supreme Court stated the principle as early as 1931 in Buck v. Jewell-LaSalle Realty Co., holding that "intention to infringe is not essential under the act." Innocence changes the price, not the verdict.

Statutory damages scale with fault, not with whether infringement occurred:

So a developer who unknowingly ships AI output that reproduced protected code can still be liable. The lack of knowledge may shrink the damages award, but it does not erase the claim. One procedural point matters before any suit: under 17 U.S.C. § 411(a), and the Supreme Court's decision in Fourth Estate Public Benefit Corp. v. Wall-Street.com, the owner of a United States work must obtain an actual copyright registration, not just file an application, before suing for infringement.

Open source is not free to take: comply with the license

"Open source" does not mean public domain. Open-source code is copyrighted, and it is released under a license that grants broad rights only if you meet its conditions. The Open Source Initiative's definition requires that these licenses allow redistribution and modification, but the grant is conditional, not a surrender of the author's copyright.

The conditions fall into two families. Permissive licenses mainly require attribution. Copyleft licenses also require that what you build and distribute stays open.

The AGPL network clause catches many software-as-a-service builders by surprise. Under section 13 of the AGPL-3.0, if you modify the program and let users interact with it remotely over a network, you must offer those users the corresponding source, even though you never distributed the code as a download.

Courts enforce these terms with real teeth. In Jacobsen v. Katzer, the Federal Circuit held that open-source license conditions run with the copyright grant, so violating them is copyright infringement, not merely breach of contract, which opens the door to statutory damages and injunctions. In Artifex Software v. Hancom, a federal court also treated the GPL as an enforceable contract. More recently, in Software Freedom Conservancy v. Vizio, a federal district court held that SFC's third-party-beneficiary breach-of-contract claim over the GPL's source-code obligation was not completely preempted by the Copyright Act, and remanded the case to state court without deciding whether SFC actually qualifies as a third-party beneficiary. The practical lesson is blunt: shipping copyleft code inside a closed, commercial product without complying can be both copyright infringement and license breach (though a contract claim over the identical conduct can be preempted by copyright law), and saying an AI tool generated it does not change who is distributing the result.

Does the AI tool or its maker share the blame?

Liability for shipping infringing output generally falls on the party who distributes it, which is usually you, not the tool vendor. Whether the AI vendor also bears responsibility is being litigated. In Doe v. GitHub, developers sued GitHub, Microsoft, and OpenAI over the Copilot coding assistant, arguing it reproduced their open-source code and stripped copyright-management information. In June 2024 the court dismissed the plaintiffs' DMCA section 1202(b) claims, while breach-of-contract and open-source-license claims survived. The court certified the DMCA ruling for interlocutory appeal that September, identifying the controlling legal question as whether sections 1202(b)(1) and (b)(3) impose an "identicality" requirement, and the Ninth Circuit heard argument in February 2026. On September 16, 2026, the panel affirmed the dismissal. It said the DMCA does not require literal identicality, but it held that the plaintiffs' output theory, that Copilot sometimes returns their code without its copyright-management information, did not state a claim because, as alleged, Copilot creates new works that never contained that information rather than removing it from copies of existing works. It declined to consider the plaintiffs' input theory, about removing that information at the training stage, because they forfeited it. That ruling addressed the DMCA claim, not whether a vendor shares liability for infringing output, and the panel expressed no view on whether the output could support a copyright-infringement claim. The breach-of-contract claims remained pending in the district court when the panel ruled.

Courts have reached different results in the related fight over training data. In Thomson Reuters v. Ross Intelligence, the Third Circuit held on September 29, 2026 that ROSS's use of copyrighted legal headnotes to train a non-generative AI for a competing product was not fair use, affirming the district court on an interlocutory appeal. The court noted that, unlike the models in Bartz, ROSS's platform "cannot generate original expression." In Bartz v. Anthropic and Kadrey v. Meta, by contrast, courts found that training large language models on books was transformative and therefore fair use, though Anthropic still faced liability for sourcing pirated copies in the first place. The Third Circuit's Ross decision concerned a non-generative tool, Bartz and Kadrey are district-court rulings, and all of these cases address training, not who owns the output.

Can you even own AI-generated code?

Often you cannot. United States copyright protects only works of human authorship. In Thaler v. Perlmutter, the D.C. Circuit affirmed in March 2025 that the Copyright Act requires a human author, and the Supreme Court denied review in 2026, leaving that holding in place. The court was careful to say this does not bar protection for work made with the assistance of AI; it bars protection for work authored by AI alone.

The Copyright Office reached the same conclusion in its January 2025 report on AI copyrightability: AI output is protectable only to the extent a human exercised creative control over the expression, and prompting alone generally does not supply that control. Its registration guidance, published at 88 Fed. Reg. 16,190, requires applicants to identify and disclaim purely AI-generated portions of a work.

The practical consequence is sharp. If your codebase was generated by an AI tool with little or no human creative input, you may hold no enforceable copyright in those portions, and a competitor could copy them freely. Yet you can still be liable if the same tool reproduced someone else's protected code. Practitioners describe this asymmetry as "all the liability, none of the protection."

Two companies, one AI, similar code: who infringes?

Probably neither, on copyright grounds alone. Copyright protects against copying, not against independent creation. In Feist Publications v. Rural Telephone Service, the Supreme Court held that originality means a work was independently created by its author rather than copied. So if Company A and Company B each prompt the same AI and receive similar code, without either copying the other, the similarity by itself does not make one liable to the other. A plaintiff must prove actual copying, typically by showing access plus substantial similarity.

The AI twist cuts deeper. If the overlapping code is purely AI-generated, it may not be copyrightable for either side, so there may be nothing to sue over. The more realistic risk in that scenario is not the two companies suing each other; it is that both pulled from a model that reproduced a third party's protected code, which exposes both of them to that third party.

How to lower your legal risk with AI coding tools

The law is unsettled at the edges, but the risk-reduction playbook is fairly settled. The following are steps organizations commonly take; they are general practices, not legal advice for your specific situation.

• Scan AI output for license provenance with software-composition-analysis tools before it ships.
• Treat copyleft matches (GPL, LGPL, AGPL) as stop-the-line events, and isolate or rewrite rather than embed them in closed code.
• Preserve copyright notices and NOTICE files for any permissively licensed components you keep.
• Document meaningful human authorship, such as selection, arrangement, and modification, if you want to claim any copyright in the result.
• Read your AI tool's terms for the scope and limits of any output indemnification it offers.
• For business-critical code, have counsel review the license posture before launch.

Next steps

The fast-moving part of this topic is who bears liability when a model reproduces its training data, and whether AI output can ever be owned. Both are in active litigation. To see how the pieces connect for United States readers, start with the AI, copyright, and code hub. If you build or sell across borders, compare the worldwide rules, where Europe in particular imposes stricter training-transparency duties. For the related question of using open-weight AI models in a product, see AI model licensing and open-source attribution.

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. Copyright disputes are fact-specific, and the law on AI-generated works is evolving, with several cited cases still pending. Consult an attorney licensed in your jurisdiction before making decisions about AI-generated or open-source code.

Related articles

• AI, copyright, and code in the United States
• 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, except the September 2026 rulings in Thomson Reuters v. Ross and Doe v. GitHub; several AI-related cases remain pending.

Frequently Asked Questions

Is code written by ChatGPT or Copilot copyrighted?

Purely AI-generated code generally is not copyrightable in the United States, because copyright requires human authorship (Thaler v. Perlmutter; U.S. Copyright Office, 2025). Code where a human made meaningful creative choices may be protected to the extent of that human contribution.

Can I get sued if an AI tool copied open-source code into my product?

Yes. Copyright infringement is strict liability under 17 U.S.C. § 501, so you can be liable even if you did not know. Innocent intent may reduce statutory damages to as low as $200 per work under § 504(c)(2), but it does not eliminate the claim.

Does 'open source' mean I can use the code however I want?

No. Open-source code is licensed copyright, not public domain. You must follow the license, which may require attribution (MIT, Apache 2.0) or that your derivative work also be released under the same open-source license (GPL, AGPL).

What is the AGPL network clause?

Under AGPL-3.0 section 13, if you modify the software and let users interact with it over a network, you must make the corresponding source available to those users, even if you never distribute the code as a download. This commonly affects software-as-a-service products.

Is copyright the same in every US state?

Yes. Copyright is exclusively federal under 17 U.S.C. § 301, so it does not vary by state. State law can still apply to related claims, such as trade-secret misappropriation or breach of a license contract.

Who is liable if Copilot suggests infringing code, me or GitHub?

Liability for shipping infringing output usually falls on the party that distributes it, which is typically the developer. Whether the tool vendor also shares liability remains unsettled; the Ninth Circuit's September 2026 ruling in Doe v. GitHub addressed only a DMCA claim.

If two companies use the same AI and get similar code, can one sue the other?

Generally no, because copyright does not bar independent creation (Feist v. Rural Telephone). A plaintiff must prove the other party actually copied its code, not merely that the two results look similar.

Can I copyright software I built mostly with AI?

Only the parts reflecting meaningful human authorship. The Copyright Office requires applicants to disclaim purely AI-generated material when registering (88 Fed. Reg. 16,190), and portions with no human creative input may be unprotectable.