GNOME contributor proposes a strict no-LLM policy for project code

A GNOME contributor has published a personal proposal on the project's blog network for how GNOME should treat code produced with large language models. The post stresses that it reflects the author's own view, not an adopted project policy.

What the proposal says

The draft policy is short: "The GNOME Project prioritizes the social and human aspects of collective software creation. Therefore: LLMs ('AI') can not be used to create or modify code submitted to GNOME, or hosted on GNOME infrastructure." It adds that a contributor may be asked to prove their code meets the requirement, and that attempts to circumvent the policy could lead to a ban.

The post also offers draft guidelines for judging a submission: contributors must be able to reason about and explain their changes, demonstrate knowledge of the problem space, fix the underlying issue rather than its symptoms, respect the time of fellow contributors, and not impersonate themselves through chatbots, agents or other automated systems.

Social norms, not workflow rules

The author sets the idea against other projects' debates. KDE has drawn attention with a controversial proposal to define an official AI policy, and Debian has voted on a similar question. In the author's view those efforts miss the point: the purpose of such statements is to shape social norms — to signal what behaviour is welcome and what is rejected — not to micro-manage how developers work.

The post deliberately sets aside the wider harms attributed to LLMs for society, workers and the environment, saying only that they run against GNOME's humanist spirit.

Why the author wants it

The argument is about people, not tooling. GNOME, the post says, is a collective that takes joy in reaching beyond individual limits, and its contributors are not payroll to be downsized. Reviewing chatbot output for free is not an attractive offer to young talent in 2026, the author writes; the project should look for the next hundred people willing to donate a little time, much as the GNOME Foundation has moved toward individual donations instead of depending on a handful of companies. The post recalls cases where companies abandoned large parts of GNOME on short notice, and argues that the project's real success metric is the community and the social bonds it creates.

Enforcement and lying

Asked how the rule could be enforced, the author answers plainly that it cannot be. People will still submit LLM-generated code, the post concedes; the policy's job is to make explicit that such careless submissions are unwelcome. The comparison is with GNOME's Code of Conduct, which is mostly about stating values and only partly about handling unwanted behaviour. On the risk of contributors lying, the post points to authorship questions in copyright: maintainers already assume, beyond a reasonable doubt, that code was written by the person who sent it, and the guidelines are a suggestion for making those judgements. The rule, the author says, targets the majority who will not try to lie at all.