COSMIC bans AI-generated code while GNOME debates whether to accept AI-made bug reports

The COSMIC desktop, developed by System76, has updated its contributor guidelines and added an explicit ban on AI-generated content. The new text requires anyone submitting a contribution to declare: "I have not included any LLM (aka AI) generated content in this PR, including code, comments, and descriptions". In other words, AI may be used to learn or to find bugs, but not to write the code, its documentation or even the pull request itself.

This stance is not unique to COSMIC. GNOME Calendar's contribution guide likewise states that "this project does not allow contributions generated by large language models (LLMs) or chatbots", and GNOME Extensions maintains a similar restriction in its review guidelines: "extensions must not be AI-generated", although it does permit AI-based learning tools and code autocompletion.

The article briefly reviews COSMIC's trajectory: the outlet reported on the project almost five years ago, which places it before the launch of ChatGPT in late 2022. Although it visually recalls GNOME —with a top panel and a dock-style launcher at the bottom— the current version of COSMIC is an entirely new development, written in Rust with the Iced toolkit, and runs exclusively on Wayland. Its first stable release shipped in Pop!_OS 24.04 in late 2025. System76's original COSMIC, by contrast, was based on GNOME with heavily customized extensions.

The piece contrasts this decision with the debate under way at GNOME, a much older project (announced in 1997, a year after KDE was announced) and a far larger one, written largely in C —a language with no automatic memory-safety guarantees— though it also uses JavaScript, C and Vala in different components. GNOME was born partly as a response to KDE, which relied on the Qt toolkit (not entirely free in its early days), prompting Red Hat to exclude KDE from Red Hat Linux and to back GNOME, written in C with GTK. Since 1999, GNOME has been the default desktop of Red Hat Linux and remains so in RHEL, and many of its developers work at Red Hat.

This is where GNOME developer Michael Catanzaro comes in. He has published an article titled "The Era of Software Quality, or the Era of Ostriches?", in which he argues that the GNOME project should accept AI-generated bug reports. It is not the first time he has made this case: back in June he published another piece, "Please Do Not Ban AI-Assisted Issue Reports", and in July he announced a reduction in the disclosure window for vulnerabilities reported to GNOME Security, which went from 90 to 30 days, effective 1 August.

Catanzaro argues that, however much effort they put in, GNOME developers will not manage to write secure code as long as they use memory-unsafe languages such as C, C++ or Vala, because it is simply too difficult even for experienced programmers. The article notes that AI is making it possible to uncover an enormous volume of flaws across free software's extensive codebases, and cites as a recent reference the 1,313 CVE identifiers listed in the latest Linux kernel security advisory in Debian.

The author of the original article does not dispute Catanzaro's arguments, but warns that this opening could be just the first step down a slippery slope, especially given the weight of Red Hat, whose leadership has been highly enthusiastic about AI (although six months later it imposed limits on developers' spending on AI tools). By this reasoning, accepting AI-generated bug reports could in time lead to accepting AI-assisted triage, then AI-assisted fixes, then AI-generated extensions, and finally core components and even the desktop itself. The text acknowledges that this would be highly controversial —as it already is in the KDE project— but does not rule out that it will end up happening in both projects, noting that Debian has already decided to allow AI contributions and that the Linux kernel itself does the same.

In short, the article portrays two opposing positions within the free desktop software ecosystem: COSMIC's, which closes the door entirely to any AI-generated content in its contributions, and that of parts of GNOME which, for security reasons and in the face of the volume of vulnerabilities AI is capable of uncovering in code written in unsafe languages, argue for opening at least a crack for bug reports, though not necessarily for the code itself.