Should AI-generated code be open source? A developer’s proposal draws pushback

Open-source developer Jakub T. Jankiewicz called on November 15, 2026, for AI-generated code to be released under open-source terms or placed in the public domain. As reported on November 17, 2026, his proposal prompted a debate over what developers owe open-source communities—and what they can keep proprietary when they use AI coding tools.

Jankiewicz argues that coding models benefit from decades of publicly available code and technical knowledge, while their outputs can be used in closed-source products. He describes that possibility as a form of “license laundering.” He also acknowledges that it can be difficult to separate human-written code from AI-generated code, particularly when developers use a model for only part of a project.

Other developers object to a blanket expectation of disclosure. One developer cited in the discussion said that choosing a product’s direction, designing its architecture, testing it and fixing errors can take months, even if an AI tool produces the code. Another developer, who said they maintain software under the AGPL, alleged that a company used a language model to clone one of their systems after a dispute over its use of their code. That account is the developer’s claim, not an established finding of infringement.

The legal questions are narrower than the proposal. In its 2025 report on copyright and AI, the U.S. Copyright Office said material generated solely by AI without sufficient human creative contribution is not protected by copyright. AI-assisted work may still qualify when a person contributes protectable expression through selection, arrangement or modification. Lack of copyright protection does not, by itself, require someone to publish source code; nor does using an AI tool automatically make its output subject to a training source’s open-source license.

Anthropic’s commercial terms say customers own outputs to the extent permitted by applicable law, and that Anthropic assigns any rights it has in those outputs to them. That arrangement does not settle whether a particular generated snippet reproduces third-party code or carries licensing obligations. Jankiewicz’s call remains a proposal, not a rule governing AI-generated software.