Should AI-Generated Code Be Open Source? The Copyright Office Backs One Premise, Not the Demand
Software / analysis
Should AI-Generated Code Be Open Source? The Copyright Office Backs One Premise, Not the Demand
A 434-point Hacker News post argues for open licences on model output; the Office's January 2025 report says prompt-only output has no human author.

Jakub T. Jankiewicz's post "All AI-generated code should be Open Source" reached 434 points and 266 comments on Hacker News as of November 15, and its argument depends on a legal premise the U.S. Copyright Office has already partly adopted.
The post makes four claims. Models were trained on open source code and Stack Exchange content, so closing the output is what Jankiewicz calls "license laundering." Code written by a model is not written by a human, so "it should not be copyrightable." Copyright rules differ by country, with the UK granting limited protection to computer-generated works. And code where humans put in little beyond a prompt "should definitely be Open Source."

What the Copyright Office has said
The U.S. Copyright Office released Part 2 of its report on AI, covering copyrightability, on January 29, 2025. The Copyright Alliance's summary says the Office found that prompting alone "does not qualify as sufficient human authorship," because users cannot control how the AI expresses and executes an idea.
That supports the post's premise. If a prompt-only output has no human author, there is no owner to apply a licence, and Jankiewicz's CC0 recommendation describes what the law may already produce.
The summary also records three limits. Iterative prompting does not change the analysis, which the Office compared to a "sweat of the brow" argument that the Feist decision rejected. Human modifications of AI output are protected, limited to the human contribution. And selection and arrangement can be protected, as in the Zarya of the Dawn registration, which covered only the "human authored original selection and arrangement" of AI images and human text.
The Office also left the door open: if future AI lets prompts closely control expression, "a different conclusion may be called for." The summary does not mention code, so none of this was written about software specifically.
Where the argument gets harder
Jankiewicz concedes the first complication himself: "code is not always 100% manual or 100% automated." A repository where a developer rewrites a model's draft contains human expression in the rewritten parts and none in the rest. Under the Office's framework, the human contribution is protected and the remainder is not, so a rule of "release AI-generated code as open source" would need a way to mark which lines are which. The post offers no figure and no mechanism for that.
A second problem is that a licence is a grant by an owner. Where nobody owns the output, an open source licence adds nothing legally, and the demand turns into a disclosure norm: say which code came from a model. That is an easier ask than the one in the headline, and it is the one a maintainer can act on.
The third is the training-data claim. The post treats the Stack Exchange material as the clearest case, citing Creative Commons attribution terms. That is a licence-compliance argument about training, which is a separate question from who owns the output, and the Copyright Office summary does not address it.
The cases the post cites
The post leans on the Monkey Selfie dispute as precedent that non-humans cannot hold copyright, and notes that the U.S. Copyright Office later referred to it regarding AI. It also points to Wikimedia Commons policy, which requires purely AI-generated images to be tagged as public domain. Commons already runs the rule Jankiewicz wants for code: no human author, no copyright, so the file goes in as public domain.
The post has no date on the page beyond a 2026 copyright footer, cites no statute number and carries no figures.
How the Office's findings map onto a repository
| Situation in a repository | Protected under the Office's findings, per the Copyright Alliance summary |
|---|---|
| Code produced from prompts alone | No |
| Prompts revised and resubmitted until the output fits | No, the Office called it "re-rolling the dice" |
| A human edits or rewrites the model's output | Yes, for the human contribution only |
| A human selects and arranges AI output | Yes, for the selection and arrangement |
The practical gap
Neither side of the argument says who checks. The post gives maintainers no method for telling model output from human edits, and the Copyright Office summary is about registration decisions, where an applicant must disclaim the AI-generated parts, as in the Rose Enigma example, where the Office granted registration after the applicant disclaimed them. A repository has no such step. Commit history is the only record most projects have, and it rarely says whether a model wrote a hunk.
What a maintainer can do today
The table points to one practical step: keep a record of which commits started as model output and which were rewritten by a person. That record is what lets a project say, line by line, where its copyright claim begins. A project such as Lonewolf, which publishes under an open licence, would need the same record if it ever had to defend a licence term, and Alibaba's OpenCodeReview is an example of an agent-adjacent tool already released as an open repository.
The dated item to watch is the Office's own caveat that its case-by-case analysis may change as tools do. Jankiewicz's post gives no date for a follow-up, and the Copyright Office summary gives none either.
Sources
More in Software
- 01Zigzag, a 35 MB Mac Browser, Courts Arc Users 35 Days Before ShutdownThe WebKit-based browser imports Arc spaces and pinned tabs, but its README lists no sync, no Boosts and a partial Chrome extension API.
- 02Nimony 0.6.2 Sets a Year-End Goal for Nim 3 With 44 Days LeftThe compiler that will become Nim 3 is MIT-licensed and builds natively without LLVM, but its FAQ says it cannot yet compile Nim 2 code unchanged.
- 03Alloy CSS Runs Menus and Tabs Without JavaScript, but Dialogs Need Safari 26.2The 24 KB MIT stylesheet leans on popover and invoker commands, so its no-script promise has browser floors that differ by component.
- 04Windows 11 Telemetry Flags Which Coding Agents You Run, Researcher FindsXusheng Li traced a service-experiment feature in build 26200.9974 that matches executable names such as claude.exe and queues the result as optional diagnostic data.