The Fork That Broke the Policy
In December 2025, Anthropic acquired Bun, the JavaScript runtime written in Zig. By April 2026, the Bun team announced a 4× compile-time improvement on their fork of the Zig compiler, achieved through "parallel semantic analysis and multiple codegen units to the llvm backend." They also announced they would not upstream the work because "Zig has a strict ban on LLM-authored contributions."
The framing landed poorly. Zig core contributors pointed out that parallel semantic analysis has been a planned feature for years, with implications for both the compiler and the language itself. The AI-ban explanation was a diplomatic way to avoid airing engineering disagreements publicly.
What the Policy Actually Says
The relevant clauses in Zig's Code of Conduct under "Strict No LLM / No AI Policy" are three:
- No LLMs for issues.
- No LLMs for pull requests.
- No LLMs for comments on the bug tracker, including translation. English is encouraged but not required.
The translation clause is the surprising one. It disambiguates the policy from a code-quality rule. A blanket ban on LLM-mediated communication, including translation, is a stance about what the project's communication channels are for.
Contributor Poker: The Core Argument
Loris Cro, Zig Software Foundation VP of Community, authored the rationale post on April 29, 2026. The argument follows three moves:
-
Empirical observation: LLM-based contributions have been mostly negative — "worthless drive-by PRs full of hallucinations," "insane 10 thousand line long first time PRs," and even PRs that looked fine but where follow-ups revealed the author was secretly consulting an LLM.
-
Normal response to overload: The project's standard answer is not to raise the quality bar but to "help new contributors to get their work in, even if they need some help getting there." This is both ethical and strategic — the primary investment is the contributor, not the patch.
- LLM-mediated contribution breaks the arithmetic: Even a perfect LLM-mediated PR means the maintainer's time was spent reviewing, not investing in a future contributor. Cro's metaphor: "In contributor poker, you bet on the contributor, not on the contents of their first PR."
Where Other Projects Landed
| Project | Stance | Mechanism | Stated Reason |
|---|---|---|---|
| Zig | Total ban on issues, PRs, comments (incl. translation) | Code of Conduct clause | Contributor cultivation |
| NetBSD | LLM-generated code presumed tainted | Commit Guidelines amendment, May 2024 | License-compatibility risk |
| Gentoo | Forbids contributions created with natural-language AI tools | Council motion, April 2024 | Copyright, quality, ethical concerns |
| curl | Bans AI-generated security reports; closed HackerOne program | Policy updates 2024–2026 | 0 valid vulnerabilities from ~20% of submissions |
| Apache | AI-assisted contributions allowed with disclosure | Generative Tooling Guidance | Pragmatic neutrality plus license clearance |
Zig's argument is the only one primarily about what reviewing is for, and the only one with a translation clause.
The 2026 Argument: Pro and Con
The HN thread on Cro's post drew 415 comments. The strongest pro-policy argument: "We do not need a middleman to talk to AI models. We are not bottlenecked by coding." If the maintainer's bottleneck is reviewing, and LLM-mediated PRs concentrate reviewing cost without distributing contributor-development benefit, the asymmetry is structural.
A counter-argument framed AI as assistive technology — comparing it to a screen reader or a robotic exoskeleton that enables contributors who otherwise couldn't participate. Cro's post acknowledges the policy will produce false negatives but accepts them because the contributor-investment problem is better served by doing so.
The Crisis-Mode Reading
One commenter noted that contributions to open-source projects were already in "borderline crisis mode" before LLMs arrived. The policy is a triage decision under a constrained reviewer budget. Another argued that the next generation of developers will grow up using AI assistance and none will become Zig contributors under a ban. Both readings can be right; the question is which becomes load-bearing first.
Coda
Zig's policy is most precisely read as a contributor-cultivation policy that forbids the input class most likely to produce contributions that don't grow contributors. The diagnostic over the next eighteen months is whether other mid-tier projects publish similarly reasoned policies or settle into vibes-based defaults. The Bun-Anthropic fork story is a small first sample of the new genre: a contribution offered, a policy invoked, a separate engineering reason left politely unspoken.
For now, if you're contributing to Zig, leave the LLM at the door. If you're a maintainer elsewhere, consider whether your review budget is investing in people or just processing patches.
