AI contribution policies compared: 21 open-source projects

What each project’s own policy says about AI-assisted contributions: its stance, what disclosure it asks for, and when it was last checked. Every row links to the project’s policy and to the catalogue entry that quotes it word for word.

How to read this table

Each summary is a short paraphrase; the catalogue entry holds the verbatim excerpt and the detail a one-line summary loses. Policies change, so every row carries the date it was verified against the project’s own source. Check the source before relying on a row. The grouping into four kinds of stance follows the catalogue’s stance codes.

All 21 sources were checked for changes on 25 September 2026. Asahi Linux, Bevy, Mesa and Rust had changed their policies since their catalogue entries were captured, and the Fedora row now cites Fedora’s adopted policy rather than the earlier proposal; those five rows describe the current policies and are dated accordingly, and their catalogue entries carry a dated update note.

Summaries paraphrase each project’s published policy for comparison; the project’s own document is authoritative. Project names identify the projects only; no affiliation or endorsement is implied.

Spot an error? Open an issue (opens in a new tab) or email ecogetaway@gmail.com (opens your email application). Corrections are made promptly and logged publicly in the repository.

What the comparison shows

• Disclosure is the common ground among permissive policies, but the mechanisms don’t match: commit trailers (Fedora, LLVM, Mesa), a PR-description statement (Kubernetes), a PR-template line (llama.cpp), a title prefix (Godot), or nothing at all (OpenStreetMap, AetherSDR).
• Autonomous agents are treated separately from AI-assisted people in several policies: Home Assistant, LLVM and Mesa prohibit them, while Godot writes rules addressed to them.
• Bans rest on different grounds: copyright and provenance (QEMU, NetBSD), legal risk to clean-room reverse engineering (Asahi Linux), or quality and review burden.

Four of these policies also say how contributors may use machine translation: what they say, quoted and compared.

The formats differ from project to project, and no tool can read them, which is why the initiative proposes a machine-readable ai-contribution-policy.yml that CI and contributor tools can read.

Related research

Academic studies of AI contribution policies in open source, for readers who want wider samples and different methods:

• AI Policy, Disclosure, and Human in the Loop: How Are Contribution Guidelines Adapting to GenAI? (opens in a new tab) (arXiv, May 2026; accepted to ICSME 2026)
• Regulating the Machine Contributor: Governance and Policy Alignment in Open Source (opens in a new tab) (arXiv, June 2026)
• “We Permit the Use of AI, but […]”: The Landscape of AI Policies in Popular Open Source Projects (opens in a new tab) (arXiv, September 2026)
• You can contribute if you… An Empirical Framework of AI Contribution Policies in OSS (opens in a new tab) (arXiv, September 2026)

This catalogue differs mainly in method: each entry is quoted from the project’s adopted text, dated, and re-checked weekly for changes.

Missing a project?

Adding a catalogue entry is a single-file pull request of about fifteen minutes, from a template. Entries must quote the project’s own policy and record the date it was checked.