A recent debate asked whether code written by artificial intelligence should be open source by default. The argument starts from a claim that AI models learned from code shared by developers. Its author asks whether output from those models should return to the open-source commons. But a moral preference, a copyright question, and a software license are different things.[1]
That difference matters to anyone shipping code from a coding assistant. The label "AI-generated" does not answer who owns the copyright, whether the code must be published, or what license applies. It also does not tell you whether the code is safe to run.
The original argument appeared on Jakub T. Jankiewicz's blog. It says code generated with little human contribution should be released under an open-source license or dedicated to the public domain. The Hacker News discussion that followed raised a second question: if code has no copyright owner, can an open-source license control how others use it at all?[1][2]
A useful way to approach this is to separate four decisions: human contribution; applicable rights; whether to publish; and how to review the code. Those choices are related, but they are not interchangeable.
Open source is a permission system, not a synonym for public
"Open source" has a specific meaning. The Open Source Initiative's definition requires access to source code and a license that allows redistribution and derived works, meaning modified versions. It also bars restrictions against particular people, groups, or fields of use.[3]
That means open source is not simply code anyone can read. A public repository without a license does not automatically grant permission to copy, modify, or redistribute the work. Conversely, software can be distributed under an open-source license while its authors retain copyright. The license grants permissions under stated conditions.[3][4]
Public domain is a different idea. In general terms, it means no copyright restrictions control the work. An open-source license depends on copyright to grant permissions and, in some cases, require conditions from redistributors. If a particular output has no copyright protection, attaching a license may not create the same enforceable conditions. The details vary by jurisdiction, meaning the country or region whose laws apply, and by the facts. This is not a legal conclusion about any specific code.
The OSI says public-domain software may be open source for most practical purposes, but it is not released under an approved open-source license. It recommends using an approved license instead of waiving copyright. The OSI does not recommend CC0 for software. The FAQ says the waiver may not work in some countries, so CC0 includes a backup license for those cases. It also notes that CC0's patent clause raised concerns during license review.[4]
There is another difference that is easy to miss: making source public and making it open source are not identical. The OSI says that source needs to be available for software to meaningfully be called open source. A repository can show code to visitors without clearly granting an approved set of reuse rights. Publication answers "can I see it?" A license answers "what may I do with it?"[4]
What current U.S. copyright guidance says
The U.S. Copyright Office's 2025 report says copyright does not extend to purely AI-generated material or material where human control over expressive elements is insufficient. Copyrightability means whether a work qualifies for copyright protection. The Office also says human authors may hold copyright over their own expression that appears in an output. Copyright may also cover creative selection, coordination, or arrangement of material and creative changes to the output. The Office treats the question as case-specific. It says prompts alone do not provide sufficient control under current generally available technology.[5]
This guidance does not say that every file touched by an AI tool is uncopyrightable. Nor does it settle whether a specific generated function contains protected human expression. A person might plan a system and choose its design. They might provide source material and edit the generated code. They might also combine separate pieces into a larger work. The human contribution can be substantial in one project and minimal in another.
The report addresses U.S. law. Teams should not apply it as a rule for every country.[5]
Training-data disputes are also separate from output licensing. The blog post argues that models learned from open-source code and that proprietary output can feel like taking value from the commons.[1] That is an ethical and policy argument. It does not by itself answer whether a specific output reuses protected expression or creates a license obligation. Those questions depend on the facts and applicable law.
The Copyright Office treats training as a separate part of its AI report. Part 3 covers generative AI training. As of November 18, 2026, the page still labels it a pre-publication version. Part 2 addresses copyrightability of AI outputs. Keeping those topics separate helps avoid treating a debate about training data as a settled rule for every generated code file.[6]
The practical problem: "AI-generated" has no clean boundary
Consider three situations:
- A developer accepts a short completion while writing a function.
- An agent drafts a module. A developer then changes its structure, tests it, and adds product-specific logic.
- A user asks for a complete application and publishes the result without review.
All three involve an AI system, but the human role differs. The Copyright Office's case-by-case approach makes that distinction relevant to copyrightability. It also makes a blanket rule difficult to apply. If "AI-generated" includes every line that received an autocomplete suggestion, a rule could cover much of modern software development. If it only applies to untouched output, teams need a way to identify which parts qualify and how that record survives edits and refactoring.[5]
A simple checkbox in a repository cannot resolve that question. Nor can a prompt transcript alone prove the human contribution to every line. Teams can keep useful records. Save the task and design notes with the generated patch. Keep human edits and test results there too. Record review decisions. These records help explain the development process. They do not guarantee copyright or prevent every claim.
The original proposal says low-effort, mostly unreviewed output should be open source. That is a reasonable policy someone may choose for their own work. But it is different from saying that current law already requires the source to be published or gives the public an automatic license. Even if a work is not protected by copyright, that does not necessarily create a duty to publish it. The HN discussion surfaced this gap between "free to use" and "must be released."[1][2]
A decision guide for teams
Before releasing code, answer these questions separately.
1. What did a person contribute?
Record what a person contributed. Did they design the system or specify its behavior? Did they select among outputs or modify the code? Keep evidence with the change record. Do not treat prompt length as a direct measure of authorship. The Copyright Office says detailed instructions alone do not necessarily control how an AI system expresses them.[5]
This record is useful even when the team does not plan to publish. It lets reviewers understand where the code came from and which parts received human attention. Do not claim that a contribution log guarantees copyright protection. It is documentation, not a ruling.
2. What license is the team choosing?
If the team controls the relevant rights, it can choose an established license that fits the project. The license defines how others may use and redistribute the code.[3][4]
If a portion may lack copyright protection, do not assume that attaching a license automatically imposes enforceable restrictions on that portion. The legal effect depends on what rights exist and which jurisdiction applies. For a commercial product or a high-value release, have qualified counsel review the intended terms.
If the code is being kept proprietary, that choice is also distinct from whether AI helped produce it. The team still needs to review model-provider terms and company policy. It should also check third-party code and contract obligations. This article does not determine those obligations for a particular tool or organization.
3. Will the source be published?
A public repository is a distribution choice. An open-source license is a permission choice. Decide both explicitly. If the team wants public review but does not want to grant broad reuse rights, get legal advice on the appropriate publication terms rather than leaving the repository unlicensed and assuming the outcome is clear.
If the team chooses an open-source release, include the license file and keep attribution and notices required by the project's dependencies. The license should be selected for the project's goals, not as an automatic consequence of having used an AI assistant.
4. Has the code been reviewed as software?
A license does not test correctness. Public visibility does not prove safety. Before shipping, review the patch and run tests. Inspect dependencies. Check for secrets or copied code. Treat generated code as code that needs verification, not as code with a special quality guarantee.
A practical review record can capture:
- The prompt or task description, where company policy allows retaining it.
- The generated patch and the final human-reviewed diff.
- Tests and static checks that ran.
- Dependency and license scan results.
- The person who approved the change and any unresolved risks.
This process cannot prove that code has no similarity to existing work. It can make review more deliberate and help teams investigate questions later.
Four possible outcomes
The same AI-assisted workflow can lead to different release choices:
- Open-source release: Publish the source with a license chosen for the project and rights the team can grant.
- Proprietary release: Keep the code private under the applicable contracts and rights, while maintaining normal review and provenance records.
- Public-domain dedication: Where appropriate, use a recognized dedication instrument after checking its effect in relevant jurisdictions. Do not confuse a dedication with an open-source license.
- No distribution: Keep the source internal. A conclusion about copyrightability does not itself require public release.
None of these choices makes code correct, secure, or free of third-party rights concerns. They answer different questions. Open source can make code easier to inspect and reuse. Proprietary code can still be reviewed internally. A public-domain dedication can reduce copyright restrictions, but it does not guarantee that the output contains no copied expression or other legal risk.
The better question is not "Was AI used?"
The debate is useful because it exposes a real tension: shared code helped build today's software ecosystem, while developers and companies still make choices about what they publish. But a blanket "AI means open source" rule merges policy, copyright, and publication into one label.
For a working team, identify the human contribution and the rights that apply. Decide what permissions to grant. Keep evidence for the release decision. Then review the code for correctness and security. Those steps remain necessary whether the final repository is public or private.
AI involvement alone does not tell you which license to choose. Make the licensing decision deliberately, document the human work, and get legal advice when the consequences are significant.
Has your team made a clear rule for code generated by AI, or is it still handled case by case?
Sources
[1] Jakub T. Jankiewicz: All AI-generated code should be Open Source
[2] Hacker News discussion of the proposal
[3] Open Source Initiative: The Open Source Definition
[4] Open Source Initiative: Frequently Asked Questions
[5] U.S. Copyright Office: Copyright and Artificial Intelligence, Part 2: Copyrightability
[6] U.S. Copyright Office: Copyright and Artificial Intelligence
Top comments (0)