MicroZig adopts a strict no-LLM policy, banning AI-written code, prose, editing and bug hunting

MicroZig, the Zig Embedded Group's open-source toolkit for building embedded applications in the Zig language, has added a blanket ban on large language models to its code of conduct. The rule, titled "Strict No LLM / No AI Policy", took effect on 28 October 2026. It was introduced through pull request #1103, opened by maintainer Matthew Knight and merged into the main branch early on 29 October (UTC) without review comments.

What the policy forbids

The text goes further than the disclosure or labelling rules many open-source projects have adopted. Contributors may not submit LLM-generated code or prose, nor paraphrase such output. Using a model to fix spelling or grammar, to translate, or to brainstorm ideas that a person then writes up is also excluded. The document adds a line that drew attention: "No LLMs for finding bugs." Issue comments that rely on chatbot advice are described as unwelcome, and talking about the use of chatbot services is itself off-limits.

Non-English speakers are told they may post in their native language and let others translate with their own tools. The code of conduct says it is largely taken from Zig's, and Zig's own page carries an almost identical policy. Enforcement falls under the general rules, with reports going to maintainers Matthew Knight and Felix Queißner.

Reactions and context

On Hacker News, Knight wrote that the project's preferences on LLMs could not coexist, so it chose to avoid them, and that he would bear no grudge against anyone forking MicroZig to build an LLM-friendly variant. Commenters were split. Some backed the maintainers' right to set terms and cited the time lost triaging unverified AI-generated reports. Others questioned how the rule can be enforced, found the bug-finding clause ambiguous, and argued that refusing AI tools for bug discovery could leave security issues undetected.

The move mirrors the Zig project, whose ban was widely discussed in April. As Simon Willison summarised at the time, Loris Cro, VP of Community at the Zig Software Foundation, framed it as an investment in people: maintainers review first contributions to build trusted contributors, which LLM-assisted work does not achieve. That stance has costs: Willison noted that Bun, the Zig-based JavaScript runtime that relies heavily on AI assistance, chose not to upstream a compiler speed-up to Zig because of the ban.

For developers using MicroZig, nothing changes in the code itself; the policy governs how they may contribute to it.