Your AI-Generated Code Might Not Be Yours

If you use GitHub Copilot, Claude, Cursor or any other AI coding assistant to write production code, the legal ownership of what you ship is less settled than your licence agreement implies. The US Copyright Office ruled in January 2025 that purely AI-generated material is not copyrightable, and that prompts alone do not provide sufficient human control to earn protection. Code you wrote with heavy AI assistance sits in an uncertain middle ground: it may be copyrightable, it may not, and no court has drawn the line for software.

The answer-first version: you probably do not own copyright in the portions of your code that an AI wrote without substantial human direction, and you may not be able to prove where the boundary lies. This does not mean someone else owns it — it may be uncopyrightable altogether, like a phone book. But your employment contract, your client agreement and your open-source licence all assume you hold full copyright in your deliverables. That assumption is now an open question.

The rule, plainly stated

Purely AI-generated output is not copyrightable in the United States. This is not a prediction or a legal opinion; it is the stated position of the US Copyright Office, set out in its January 2025 report Copyright and Artificial Intelligence, Part 2: Copyrightability. The report received over 10,000 public comments and represents the Office’s most comprehensive statement on the subject. Its conclusions are clear: “material generated wholly by AI is not copyrightable”, and existing law is adequate to handle the question without new legislation.

The nuance sits in the middle ground — which is exactly where most AI-assisted coding lives.

What the Copyright Office said

The report draws several lines. First, it confirms the long-standing requirement that copyright requires a human author. An AI system cannot be an author, regardless of how sophisticated its output. Second, it addresses prompts: “based on the functioning of current generally available technology, prompts do not alone provide sufficient control” to constitute authorship. Telling an AI what to write is not the same as writing it.

Third — and this is where it gets useful — the report recognises that human contributions to AI-generated output can earn copyright. “Human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs, as well as the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.” In plain language: if you substantially modified, selected among or creatively arranged AI-generated material, those contributions are protectable. The AI-generated material itself is not.

The catch is that this is assessed case by case. There is no bright-line rule, no percentage threshold, no checklist. Whether your contribution is “sufficient” depends on the specifics — and for code written with AI tab-completion tools, nobody has tested where the line falls.

The code-specific gap

The Copyright Office report addresses “works created using generative AI” broadly — text, images, music, video. It does not specifically analyse software code. This matters because code has properties that make the authorship question both harder and less intuitive than for prose or art.

Consider the daily workflow of a developer using Copilot or Cursor. The tool suggests a function body; the developer accepts it, changes two variable names and adds a comment. The tool suggests an import statement; the developer accepts it verbatim. The tool suggests a test; the developer rewrites half of it. Across a morning’s work, the code that ships is a patchwork of human-written, AI-generated and human-modified-AI-generated material, and the boundaries between them are not recorded anywhere.

Unlike a novelist who asks an AI to draft a paragraph and then rewrites it — where the creative transformation is at least arguable — a developer who accepts a syntactically correct function verbatim has contributed what, exactly? The prompt (“write a function that sorts by date”), the context (the surrounding code the model used as input), and the acceptance decision. The Copyright Office says prompts alone are not enough. Whether the acceptance-and-integration of a code suggestion constitutes the kind of “creative selection” the report describes is untested. We have written before about the productivity claims around AI coding tools; the ownership question is their unreported underside.

Thaler and the authorship line

The leading US case is Thaler v. Perlmutter, decided by the DC District Court in August 2023. Stephen Thaler attempted to register a work of visual art created autonomously by his AI system DABUS, listing the AI as the author. Judge Beryl Howell ruled that copyright requires human authorship and that an AI system cannot be an author: “Copyright has never stretched so far as to protect works generated by new forms of technology operating absent any guiding human hand.”

The case is clear on its facts — a fully autonomous AI creation with no human creative input cannot be copyrighted. But it does not resolve the harder question: how much human direction, modification or selection turns AI-assisted output into a protectable work. For developers, this is the question that matters. Nobody is shipping code that an AI wrote with zero human involvement; the real workflow is a continuum of human and machine contribution, and the law has not yet said where on that continuum protection begins.

The Copilot litigation

The other live proceeding is Doe v. GitHub, the class action filed in November 2022 in the Northern District of California. The plaintiffs allege that GitHub Copilot reproduces copyrighted open-source code — including code released under licences like GPL that require attribution and licence preservation — without complying with those requirements. The case addresses the input side of the copyright question (whether training on copyrighted code is permissible) rather than the output side (whether what Copilot generates is copyrightable). But it highlights a practical risk.

If Copilot can and does reproduce fragments of GPL-licensed code in your commercial project, the copyright status of both the fragment and your surrounding code becomes complicated. You may be shipping code that carries licence obligations you did not intend to accept, embedded in a codebase whose own copyright status is uncertain. This is not a theoretical scenario; researchers have documented cases of Copilot suggesting code that closely matches existing open-source implementations. The combination of uncertain output copyright and potentially encumbered training inputs makes the legal surface area larger than most developers realise.

What the EU says

The EU has not unified its position on AI output copyrightability. The EU AI Act, which phased into force over 2025 and 2026, addresses transparency obligations and training-data requirements but does not directly adjudicate whether AI-generated content is copyrightable. That question remains with individual member states’ copyright frameworks and, ultimately, the Court of Justice of the European Union, which has not ruled on AI authorship.

The UK has an unusual provision: Section 9(3) of the Copyright, Designs and Patents Act 1988 provides for copyright in “computer-generated” works, with the author being “the person by whom the arrangements necessary for the creation of the work are undertaken.” This was written decades before generative AI and its applicability to modern AI coding tools is untested. Some commentators argue it could protect AI-assisted code; others argue the provision was aimed at much simpler computer processes. Until a court considers it in the context of a Copilot-like tool, it remains an interesting footnote rather than a reliable basis for planning.

For developers working across jurisdictions — which, given the nature of open-source, is most of them — the patchwork of national approaches adds a layer of uncertainty. We have explored the training-data copyright question before; the output-ownership question is its mirror image, and it is no better resolved.

What it means for the code you ship

Several practical consequences follow, even though the courts have not yet ruled definitively:

• Your employment contract probably assumes full copyright. Most developer employment agreements include an assignment clause: the company owns the intellectual property in everything you create in the course of your work. If 40% of that work was generated by an AI and is potentially uncopyrightable, the assignment clause is promising something you may not have to give.
• Open-source contributions carry a copyright warranty. When you submit code to an open-source project under a licence like Apache 2.0 or MIT, you are implicitly representing that you have the right to licence it. If a substantial portion was AI-generated and uncopyrightable, that representation may be more fragile than you think.
• Client deliverables may have a gap. If you are a freelancer or consultancy delivering code to a client with a copyright assignment, the portions the AI wrote may not be assignable — not because someone else owns them, but because nobody does.
• The practical risk is low today. No court has yet refused to enforce copyright on AI-assisted code, no company has lost a case because Copilot wrote 30% of a module, and no open-source project has been challenged on these grounds. The legal uncertainty is real; the practical consequences have not yet arrived.

The prudent response is boring and effective: keep records of what you wrote versus what the AI generated. Review and substantially modify AI suggestions rather than accepting them verbatim — not only because the code is likely better for it, but because your creative contribution to the final result is what earns copyright protection. Be cautious about representing to clients or in open-source contributions that you hold full copyright in heavily AI-assisted code. And pay attention to the Doe v. GitHub case, because its outcome will shape the practical landscape more than any amount of commentary.

The hallucination problem taught developers that AI output needs checking for correctness. The copyright question teaches a parallel lesson: AI output needs checking for ownership. The tool is useful. The assumption that you own what it produces is the part that has not been tested.