Short answer: There are two separate questions. (1) Can you own AI-generated code? The US Copyright Office’s 2025 report concluded that purely AI-generated output is not copyrightable without sufficient human authorship, judged case by case. (2) Can AI-generated code infringe someone else’s? That is the subject of the GitHub Copilot litigation (Doe v. GitHub), where a court dismissed most claims and the matter went to appeal. Neither question has a simple universal answer, so teams should govern authorship, provenance, and license review. This article is general information, not legal advice.
“Who owns the code the AI wrote?” sounds like a philosophical question. For an engineering organization it is a practical one, with two distinct halves that are easy to confuse: whether you can protect AI-generated code, and whether it might infringe someone else’s rights. Recent developments give useful signal on both—without settling either.
Question 1: Can AI-generated code be copyrighted?
In its 2025 report on copyright and artificial intelligence (Part 2, on copyrightability), the U.S. Copyright Office addressed when AI outputs can be protected. Its conclusion, as widely summarized, is that outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements—and that this must be assessed case by case.
In plain terms:
- Code produced purely by prompting an AI, with no meaningful human expressive contribution, is unlikely to be copyrightable.
- Human selection, arrangement, and modification of AI output can support protection for those human contributions.
- The dividing line is fact-specific: how much did a person shape the expressive result?
For a team, the practical implication is that heavy, unreviewed reliance on generated code may weaken your copyright position in that code—another reason human authorship and review are not just quality controls but IP controls.
Question 2: Can AI-generated code infringe others’ code?
This is the question behind the GitHub Copilot litigation (Doe v. GitHub). Plaintiffs alleged that Copilot could reproduce publicly licensed code without honoring its license terms or attribution. As widely reported, the court dismissed the majority of the claims—including a Digital Millennium Copyright Act §1202(b) claim—with some claims dismissed with prejudice, and the dispute proceeded to appeal before the Ninth Circuit.
Two honest caveats:
- This is ongoing litigation whose outcome is not final; appellate decisions can change the picture.
- A dismissal of specific claims is not a blanket ruling that AI code generation never raises IP issues; it is a ruling on particular legal theories and facts.
The signal for teams is risk management, not panic: generated code can resemble training data, and license obligations (including copyleft terms like the AGPL network-use clause) attach to code regardless of how it was produced.
Two questions, one table
| Question | Recent signal | What it does not settle |
|---|---|---|
| Can you own AI-generated code? | USCO (2025): protectable only with sufficient human authorship | The exact threshold; it is case-by-case |
| Can AI code infringe others’? | Copilot litigation: most claims dismissed, on appeal | Whether outputs can ever infringe; case is ongoing |
A governance checklist that does not require a lawyer to start
You will still want qualified counsel for specifics, but engineering can reduce risk today:
- Keep humans in authorship. Require meaningful review and modification of generated code, both for quality and to strengthen any copyright position.
- Track provenance. Record where significant AI assistance was used, so ownership and licensing questions can be answered later.
- Review licenses of suggested code. Treat AI suggestions like any third-party code: check for recognizable licensed snippets and obligations.
- Mind copyleft. Understand how licenses like AGPL-3.0 apply to what you ship; see the license guide and the AGPL obligation navigator.
- Prefer tools you can govern. A platform that records tasks and reviews makes provenance and oversight enforceable rather than anecdotal.
Where a managed platform helps
Copyright and IP hygiene depend on evidence: who reviewed what, and where AI was involved. Ad-hoc use on personal machines leaves no trail. A platform that runs AI work through recorded tasks and a review workflow—the model MonkeyCode describes—makes that evidence a byproduct of normal work. It does not answer the legal questions for you, but it gives your counsel something to work with.
Bottom line
The ownership question and the infringement question are different, and both currently resolve to “it depends.” The US Copyright Office ties protectability to human authorship; the Copilot litigation shows the infringement debate is live but far from a sweeping rule. The durable response is governance: keep humans authoring and reviewing, track provenance, respect licenses, and use tools that make that discipline enforceable.
Source boundary: The copyrightability conclusions are summarized from the U.S. Copyright Office’s 2025 report on copyright and AI (Part 2), and the litigation summary from public reporting on Doe v. GitHub, both checked 20 July 2026. Legal matters evolve and appellate outcomes may change this picture. This article is general information, not legal advice; consult qualified counsel for your jurisdiction and facts.