AI Outlawed ☠️ in Open Source Project

A GPL-licensed open-source project is tightening its contribution rules by refusing patches that use AI code generators, citing unresolved—and potentially incompatible—copyright and licensing questions around AI-generated code. The move matters because the project’s GNU GPL v2 “copyleft” terms can require downstream recipients to adopt the same license, so uncertainty about whether AI outputs carry or reproduce material from differently licensed sources becomes a direct compliance risk.

The policy centers on the idea that, for AI-assisted code, the copyright and license status of the output is “ill-defined” and lacks a settled legal foundation. Even when training data is known to include open-source material, the training mix may include code under restrictive terms that aren’t compatible with the project’s licensing requirements. That uncertainty collides with the project’s contribution mechanism: contributors must certify compliance through a DCO (Developer Certificate of Origin) process, which requires submitters to understand the copyright and license status of what they contribute. If contributors can’t credibly determine whether AI-generated code meets those licensing obligations, the certification becomes hard to justify.

The project’s stance is framed as risk management rather than a blanket anti-AI position. It draws a boundary around what counts as “AI code generators” in the contribution itself, while allowing other AI uses—such as researching APIs or algorithms, static analysis, and debugging—so long as the resulting output is not included in the submitted patch. Tools explicitly named as impacted include GitHub Copilot, along with other AI code generators and agent-style systems built on similar tooling.

The policy also reflects the project’s “viral” licensing reality. With GPL v2, incorporating or distributing derivative works can trigger license obligations for the broader product or distribution. That makes the project less tolerant of ambiguity than permissively licensed projects, because a licensing mismatch could force legal or compliance fallout for anyone who ships the resulting code.

In the surrounding discussion, commenters argue over enforceability and detectability. Several contend that it’s often impossible to tell whether a specific patch was produced with an LLM, making the rule difficult to police in practice. Others respond that the policy is still meaningful as a contributor standard: even if verification is imperfect, the project can refuse contributions where AI usage is known or suspected, and it can rely on the DCO certification as the accountability lever.

The debate spills into broader frustrations about open-source governance and internet culture, but the core issue stays technical and legal: until courts and regulators provide clearer guidance on how training data licenses apply to AI outputs, GPL-based projects may treat AI-generated contributions as too uncertain to certify. The policy is positioned as strict for now, with an explicit openness to revision as legal interpretations mature.