GCC bars LLM-generated code from legally significant contributions

The GCC steering committee has adopted an AI contributions policy drafted by its dedicated AI policy working group. The core rule: the compiler project will refuse any legally significant contribution that contains or is derived from LLM-generated content. “Legally significant” borrows the GNU Project maintainer standard, which sets the bar at roughly 15 lines of code or text — the point at which a contribution starts to matter for copyright purposes. One notable carve-out lets maintainers accept LLM-generated test cases at their discretion.

The policy targets what ends up in the codebase, not how developers work. Using an LLM for research, code analysis, bug hunting, reporting, or reviewing patches remains fine, so long as the model’s output doesn’t make its way into the actual contribution. The committee has framed the rule as provisional, saying it expects to revisit and revise it as the legal and technical picture shifts.

The move reflects the copyright and provenance anxieties now rippling through major open-source projects, where the unsettled legal status of machine-generated code is a real liability for a foundational toolchain like GCC. It lands GCC alongside other communities drawing explicit lines around AI-authored contributions, favoring a cautious, copyright-driven stance over an outright ban on AI tooling.