Chezmoi introduces ban on LLM-generated contributions

Original Article ↗ Hacker News Discussion ↗

Open source tool chezmoi has adopted a zero‑tolerance policy on AI assistance, banning contributors who submit any LLM‑generated content, even if reviewed. Commenters weigh the maintainers’ frustration with low‑quality “slop” and spammy PRs against concerns that such a blanket, unenforceable rule is hostile to good‑faith contributors and blurs lines around tools like Copilot. The thread also touches on unresolved legal questions about copyright and AI‑generated code, which some see as an additional motivation for strict policies.

Policy change and scope

  • Thread clarifies that the current policy is a blanket ban: any contribution containing LLM‑generated content leads to immediate ban, without recourse.
  • Earlier, more permissive language about “unreviewed” LLM content was removed; several commenters initially misread the diff and confused old vs new text.
  • Some interpret “any LLM use” narrowly (only generated content), others more broadly (even using Copilot/tab‑complete or LLMs for review could technically violate it).

Enforcement and ambiguity

  • Many doubt enforceability: it’s impossible to prove no LLM was used, and AI detectors are unreliable.
  • Others say enforcement will be social: if maintainers think something “looks like” LLM output, they’ll reject it and ban the contributor.
  • Concern is raised over false positives and no‑recourse bans for humans who just wrote bad or unfamiliar code.

Maintainer motivations and experience

  • Commenters assume the maintainer is reacting to floods of low‑effort, incorrect “slop” PRs and even bogus vulnerability reports obviously produced by LLMs.
  • The linked discussion shows frustration: past attempts at “LLM allowed if carefully reviewed and declared” were ignored, leading to the hard ban.

Community impact and fairness

  • Some see the “immediately banned without recourse” language as hostile and off‑putting; they say they wouldn’t contribute under such a policy.
  • Others argue that’s the point: the project prefers fewer contributors over spending time triaging AI‑generated junk.
  • One view: the rule is mainly a cudgel to quickly eject net‑negative contributors, not a literal anti‑Copilot witch hunt for good PRs.

Alternative approaches suggested

  • Proposals include:
    • Ban only “unreviewed” or “low‑quality” LLM contributions.
    • Require disclosure of LLM use and prompts.
    • Provide project‑specific LLM contribution guidelines.
  • Supporters of the ban counter that debating “quality” is more contentious and time‑consuming than a bright‑line no‑LLM rule.

Legal and copyright considerations

  • Several comments raise unresolved questions about whether AI‑generated code is copyrightable and whether it risks “public domain contamination” of projects.
  • Others summarize recent US copyright guidance: pure AI output isn’t protected; human‑modified output might be, depending on the degree of human authorship.
  • A few speculate that a clear no‑LLM policy might be a defensive move against future legal uncertainty.