FreeBSD developer survey shows most want an AI code policy, few want a ban

The FreeBSD project has published its third status report of 2026, and it shows the community is still working out how to treat code produced with AI coding assistants. The Register reported on the document, which also covers a wide range of other development work.

The report contains 51 entries from five teams covering 16 projects, along with six further sections with more detailed write-ups. Among the items are progress on Sylve, a graphical tool for managing FreeBSD virtualization hosts, and AppJail, a containerized application distribution system that is meant to be broadly comparable to Snap, Flatpak and AppImage on Linux.

The most closely watched part for AI coverage is a subsection of the Core Team update on a developer survey about AI. Baptiste Daroussin created the survey, which drew 140 responses. According to The Register, the project first discussed the question publicly in September 2025 and has not yet issued a final answer.

On policy, 88 percent of respondents want the project to adopt one. Of those, 49 percent favor a formal written policy and 39 percent prefer light guidance and best practices. Six percent said AI use should be explicitly discouraged, and three percent said existing contribution rules are enough.

Asked how AI-assisted code should be handled, 35 percent said it must be disclosed and 25 percent said it should get stricter review than human-written code. Another 22 percent said it should be treated like any other contribution, and 16 percent said it should be rejected outright.

Respondents most often asked for mandatory human review (78 percent) and mandatory disclosure (69 percent) as safeguards. Documentation of provenance or licensing was requested by 47 percent, and mandatory testing by 39 percent. On quality, the leading worries were correctness and subtle bugs at 39 percent, licensing and provenance at 26 percent, and maintainability at 21 percent. Overall, 75 percent said they are somewhat or very concerned about copyright and licensing.

Accountability, meaning that a human stays responsible for submitted code, was the principle chosen most often at 91 percent, followed by transparency at 78 percent. Preserving code quality was the top-ranked goal for any policy. For other uses, 65 percent would allow AI-generated documentation with human review. A majority of 54 percent would permit AI tools on project infrastructure, most of them with restrictions, while 29 percent would not.

The Register contrasted the situation with other projects. NetBSD banned AI-generated code in 2024, and OpenBSD's 8.0 release materials poked fun at AI users. The outlet also pointed to a Debian developer poll it covered in August, where the outcome allowed the use of AI tools.