Who Owns AI-Generated Code? The Open Source Debate Hitting Indie Hackers
Who owns AI-generated code? On 15 November 2026 that question was the top of Hacker News. Jakub T. Jankiewicz, who posts as jcubic, submitted his essay “All AI-generated code should be Open Source” at 14:46 UTC. As of 16 November the thread had more than 900 points and 500 comments, and it had reached number one.
His answer is that the code should be open. The comments do not line up behind him, or behind each other. A January 2025 report from the US Copyright Office draws a third line, about human control. Indie hackers are already shipping products built with these tools. Shipping fast with AI made a first version cheap. It did not decide who owns the file.
A short timeline
The public record is a short calendar.
| When | What is on the record |
|---|---|
| January 2025 | US Copyright Office publishes Part 2, on copyrightability. |
| 1 January 2026 | OpenAI's terms assign the user any right the company has in Output. |
| 2 March 2026 | The US Supreme Court declines to hear Thaler v. Perlmutter. |
| 15 November 2026 | Jankiewicz's essay reaches number one on Hacker News. |
The report is Copyright and Artificial Intelligence, Part 2: Copyrightability. The terms are OpenAI's Terms of Use, effective 1 January 2026. The denial is reported by Reuters and by Mayer Brown. The essay is Jankiewicz's post. The thread is Hacker News, submitted by jcubic at 14:46 UTC on 15 November.
The case for opening the code
Jankiewicz opens by marking the claim as a hot take: “This may be a hot take, but I strongly believe that all code generated by AI should be Open Source.”
The reason he gives is the training set. He argues that the models were trained on open source code, and on Stack Overflow and Stack Exchange content published under Creative Commons, where attribution is required. He writes: “Turning this collective knowledge into proprietary, closed-source code feels like license laundering.”
On copyright, he argues that purely AI-generated code should not be copyrightable. He cites the Monkey Selfie dispute. He notes that Wikimedia Commons tags purely AI-generated images as public domain. He also notes that the United Kingdom grants limited copyright protection to computer-generated works. The essay is already treating the rule as something that changes by country.
He proposes releasing that code under an open source licence, or under CC0. The effort line is explicit. Code where “humans put little effort, except writing prompts, should definitely be Open Source.”
Should AI-generated code be open source?
The thread treated the proposal as the claim to test.
HN user ubercow13 wrote: “If AI generated code is not copyrightable, releasing it under an open source license is meaningless.” HN user glitchc made the same cut: “licenses exist because code is copyrightable in the first place”, and “If AI code is not copyrightable, then no license can govern it.” Their point sits inside the essay’s premise. If the code is not copyrightable, an open source licence has nothing to bind.
HN user delichon turned the training argument around on people: “I too have trained on the massive corpus of open source code. Should everything I code also be open source?”
HN user viraptor offered his own reading of two cases. He wrote that Doe v. GitHub found LLM output “doesn’t inherit the oss copyright notice”, and that Thaler v. Perlmutter “showed 100% AI output can’t be given any copyright”. Those are his compressions of the cases. They stay labelled as his reading.
HN user anself counted the hours that sit after the prompt: “20 mins of agentic software coding against hours or days of testing”, and then “Forcing it to be open source steals all that time and effort for me”.
HN user rvz wrote: “The author needs to realize that this is unenforceable.” HN user sorrywhat wrote about where the software might go: “we may see more software going in a source-available direction – either that or full cloud.”
HN user jph releases his AI-generated code under a multiple-choice licence: Apache, BSD, GPL, or MIT. The example he pointed at is publicprojects-io/axa.
Off Hacker News, on the same date, game designer Lars Doucet wrote that open-sourcing AI code is not yet a dogmatic principle for him, but “it’s where I’ve landed in *practice* in my daily work.” The post describes where his own daily work has landed.
Some builders have made that choice public on a product. Marc Lou open-sourced MailCheap, a vibe-coded mailing platform. BenchUp, the open source SketchUp alternative posted as KetchUp, went out as a public repo. Each one is a licence chosen for that project.
Can AI-generated code be copyrighted?
The clearest official text in this record is the US Copyright Office report from January 2025, Copyright and Artificial Intelligence, Part 2: Copyrightability.
“Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.” The prompt-only case is cut off in a shorter sentence: “prompts do not alone provide sufficient control.”
The report still leaves room for a person who used a tool. Where the tools “assist rather than stand in for human creativity”, it says that use “does not affect the availability of copyright protection for the output.” What the human can hold is named directly. A person can hold copyright in the “creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.”
On 2 March 2026 the US Supreme Court declined to hear Thaler v. Perlmutter. Reuters reported the denial. Mayer Brown summarised the same outcome. The denial left the D.C. Circuit’s human-authorship ruling in place.
A vendor’s contract cannot create a copyright the law does not grant. OpenAI’s Terms of Use, effective 1 January 2026, say: “We hereby assign to you all our right, title, and interest, if any, in and to Output.” The same terms say “output may not be unique and other users may receive similar output”. The company hands over whatever right it has. Similar output can reach another user.
Who owns AI-generated code when a person did the work?
Under that US line, a prompt alone does not supply the control the Office requires. Where copyright exists, it sits in the human contribution the report named: selection, coordination, arrangement, modification. HN user wolttam said the builder’s version of that split: “it’s the person that owns the project, and the LLM is a tool.”
That is a reading of US materials. Jankiewicz points at a different rule in the United Kingdom, which he says grants limited copyright protection to computer-generated works. The same file can fall under a different rule in another country. This is not legal advice.
Is AI-generated code plagiarism?
Copying risk exists if the output reproduces existing code near-verbatim. Check the licence on anything you recognise and keep. Generation by itself does not prove the lines were copied. It also does not prove they are original.
Jankiewicz’s training claim is the background worry, not a verdict on a particular file. He argues the models learned from open source repositories and from Stack Overflow and Stack Exchange posts under Creative Commons that require attribution. A generated file does not arrive with those notices filled in.
HN user viraptor’s reading of Doe v. GitHub is the one already quoted: the output “doesn’t inherit the oss copyright notice”. Keep it as his reading. A missing notice leaves the origin unchecked. If a block looks like a known implementation, the next step is the same as if a person had pasted it. Identify the work. Read the licence. Do not treat a clean compile as a clearance.
The sources above do not contain a plagiarism judgment about a named indie product. The Monkey Selfie citation in the essay is an authorship comparison. The Copyright Office sentences address whether copyright arises.
Can you sell an app you built with AI?
You can sell the app. Owning a copyright in every generated line is a separate question, and the Office’s report is about that second question. People are already charging. Payment-verified revenue from AI-built apps is its own record. How indie hackers get paid still runs through subscriptions, one-time sales, and the other ordinary models. None of the documents cited here say the sale is void because a model wrote a function.
What you can point to, if someone challenges the expressive part, is the human work the report describes. Keep records of the design and of the edits you actually made: the selection, the arrangement, the modifications. That is a practical consequence of those sentences. It is not a form the Office published in the passages quoted here.
OpenAI’s assignment gives you the company’s rights, if any. The same terms warn that other users may receive similar output. A moat made only of the generated source is a thin moat. Whether vibe coding is killing indie hacking is this site’s look at that pressure: a first version got cheap, and so did a copy. For someone shipping an AI-built product, the harder piece to hand to the next prompt is the product, the distribution, the data, and the support. Pieter Levels replacing paid SaaS with services he generated is a cost story about the running product he kept.
For a real dispute, talk to a lawyer. The tool’s terms, the country, and the amount of human expression can each move the answer. This is not legal advice.
Would you open-source your AI-written code, or does the human work make it yours?
FAQ
Can AI-generated code be copyrighted?
The US Copyright Office report Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025) says copyright does not extend to purely AI-generated material, or to material where there is insufficient human control over the expressive elements, and that prompts do not alone provide sufficient control. The same report says humans can hold copyright in the creative selection, coordination, or arrangement of material in the outputs, or in creative modifications of the outputs. Using AI to assist human creativity, rather than to stand in for it, does not by itself remove protection. On 2 March 2026 the US Supreme Court declined to hear Thaler v. Perlmutter, leaving the D.C. Circuit's human-authorship ruling in place. OpenAI's terms, effective 1 January 2026, assign you the company's right in Output, if any, and say output may not be unique. A vendor cannot grant a copyright the law does not. Law differs by country. The essay notes that the UK grants limited copyright protection to computer-generated works. This is not legal advice.
Should AI-generated code be open source?
Jakub T. Jankiewicz argues yes. His essay says all code generated by AI should be open source, calls closed-source use of that output license laundering, and says code where humans put little effort except writing prompts should definitely be open source. He proposes an open source licence or CC0. On the 15 November 2026 Hacker News thread, ubercow13 wrote that if AI-generated code is not copyrightable, releasing it under an open source licence is meaningless, and glitchc wrote that if AI code is not copyrightable, no licence can govern it. anself wrote that forcing it open source steals the testing time. rvz called the proposal unenforceable. Lars Doucet wrote the same day that open-sourcing AI code is not yet a dogma for him, but that it is where he has landed in practice. The law cited here does not impose the licence the essay wants.
Can you sell an app you built with AI?
Selling the product is a separate question from owning a copyright in every generated line. OpenAI's terms assign you whatever right the company has in the output, if any. The US Copyright Office report says protection tracks human creative contribution, including selection, arrangement, and modification, and that prompts alone do not provide sufficient control. People are already charging for AI-built apps. Keep a record of your own design and editing if you may need to show the human part. Another user of the same tool may receive similar output, so a moat made only of the generated source is thin. The harder-to-copy pieces are the product, distribution, data, and support. For a real dispute, talk to a lawyer. This is not legal advice.
Is AI-generated code plagiarism?
Copying risk exists if the output reproduces existing code near-verbatim. Check the licence on code you keep. Jakub T. Jankiewicz argues that models were trained on open source code and on Stack Overflow and Stack Exchange content under Creative Commons, where attribution is required. HN user viraptor's reading of Doe v. GitHub is that LLM output does not inherit the open source copyright notice. That is his reading of the case, not a holding restated here as settled law. A generated file can still match an existing work. Generation alone does not prove the lines are original, and it does not prove they were copied. This is not legal advice.
Who owns AI-generated code?
Under the US Copyright Office's January 2025 report, purely AI-generated material is outside copyright, and a prompt alone is not enough human control. A person can hold copyright in creative selection, coordination, arrangement, or modification of the output. On Hacker News, wolttam wrote that the person owns the project and the LLM is a tool. OpenAI assigns you its right in the output, if any, which does not create a copyright the law withholds. The essay's UK example is limited copyright for computer-generated works, so the same file can sit under a different rule in another country. This is not legal advice. For a real dispute, talk to a lawyer.