System76’s COSMIC desktop environment has tightened its contribution rules by banning LLM-generated content from pull requests, including code, comments and descriptions. The new requirement appears directly in COSMIC’s pull request template, where contributors must confirm that their submission does not contain AI-generated material. System76 developer Jeremy Soller said the main reason for the change is the review burden created by a growing number of AI-assisted submissions from first-time contributors.
Under the revised rules, contributors must certify that they have not included any LLM-generated content in a pull request and that they understand the changes they are submitting. They are also expected to be able to respond to review comments, accurately describe their changes, test the contribution and certify it under the project’s Developer Certificate of Origin. COSMIC says pull requests that do not meet the template requirements may be closed.
Soller explained that COSMIC has been receiving substantially more changes from first-time contributors using LLMs, but those submissions are often not aligned with planned development work and have had a relatively low acceptance rate. That creates a disproportionate workload for maintainers because every contribution still has to be reviewed, tested and evaluated by a human developer. The issue is therefore not simply whether AI-generated code works, but whether the cost of reviewing it is justified for the project.
COSMIC had already taken a cautious approach to AI-generated contributions before adopting the stricter rule. In an August discussion about the project’s AI policy, maintainers pointed users toward an existing pull request policy and indicated that large AI-written submissions could be rejected over concerns including maintainability and copyright. The new rule goes further by removing the disclosure-based approach and instead prohibiting LLM-generated material from submissions altogether.
There is at least one exception. The policy does not apply in the same way to cosmic-flatpak, where individual upstream projects manage their own Flatpak manifests and COSMIC’s developers mainly review those packages for appropriate sandboxing. That distinction reflects the different level of ownership and maintenance responsibility involved compared with changes to COSMIC’s own codebase.
The decision places COSMIC among a growing number of open-source projects trying to define stricter boundaries around AI-assisted contributions. Similar debates have emerged elsewhere over whether LLM-generated code increases productivity or simply transfers more work to maintainers who must validate unfamiliar submissions. KDE recently locked a heated discussion over proposed LLM contribution rules after community disagreement over how such submissions should be handled.
COSMIC’s policy is notable because it targets more than just source code. AI-generated pull request descriptions and comments are also prohibited, suggesting that the project wants contributors to personally understand and explain what they are submitting rather than using an LLM to mediate the entire contribution process. That requirement addresses one of the recurring concerns raised by maintainers: a contributor may be able to generate a plausible patch quickly without being able to explain its design or respond effectively when reviewers identify problems.
The move does not amount to a blanket rejection of AI tools in every development context. Soller’s explanation focused specifically on the maintenance cost of reviewing externally submitted LLM-generated contributions, rather than arguing that AI has no legitimate use in programming. The policy therefore draws a line around what COSMIC will accept into its repositories rather than dictating how developers may use AI privately for tasks such as research, debugging or experimentation.
That distinction reflects a wider tension in open-source software. Generative tools have lowered the barrier to producing code quickly, but they have also made it easier to submit large amounts of work that still require careful human review before they can be merged safely. For a project such as COSMIC, which is being actively developed by a relatively focused team, reducing low-value review work can directly affect how much time maintainers have available for planned features, bug fixes and releases.
For now, COSMIC’s position is straightforward: pull requests must be written, understood and supported by the contributor submitting them, without LLM-generated code, comments or descriptions. As more open-source projects encounter similar pressure from AI-assisted submissions, COSMIC’s decision may become one of several models maintainers use to control review workload while preserving confidence in incoming contributions.
