March 27, 2026
When bots code, lawyers reload
Vibe-Coded Ext4 for OpenBSD
AI-built Linux disk format hits OpenBSD, cue legal panic
TLDR: An AI-written ext4 driver was proposed for OpenBSD, but devs balked over who owns the copyright and whether they can legally ship it. Commenters cracked jokes about 'vibe coding,' argued that no-copyright means free to share, and even begged for an AI 'copyright wash' for Windows.
An engineer dropped an AI-made version of ext4 — a popular Linux disk format — into OpenBSD’s inbox, and the internet lit up like a server room at 3 a.m. Thomas de Grivel bragged it was built with ChatGPT and Claude, offers full read/write, passes the checker, but skips a safety feature called journaling. On the OpenBSD side, big names like Theo de Raadt and Damien Miller basically said: cool story, but who actually owns this thing? Without a clear human copyright holder, they can’t legally ship it. Cue the lawyer thunder.
The crowd response? Spicy. One user roasted the paywall + “vibe-coded” combo as peak 2024, while another declared “vibe coding and OpenBSD” the crossover no one asked for. Others were stunned that the debate is almost entirely legal, not about whether the code even works. The hottest take came from someone arguing that if AI output isn’t protected by copyright, then it’s free to distribute — a direct clash with OpenBSD’s cautious stance. And yes, someone yelled “can AI copyright-wash Windows already?!” because of course they did.
Bottom line: devs are stuck between AI magic and legal chaos, and the comments turned it into a courtroom drama with memes. Read the LWN piece here and the mailing list chatter here.
Key Points
- •An ext4 filesystem implementation was submitted to OpenBSD on March 17, claiming read/write support and e2fsck compatibility, but lacking journaling.
- •The submitter, Thomas de Grivel, disclosed the code was generated using ChatGPT and Claude-code, with his reviews and testing.
- •Concerns were raised that the code could be derivative of GPL-licensed Linux ext4 code or documentation, risking license contamination.
- •Theo de Raadt noted reimplementation for interoperability is allowed, but emphasized that unclear copyright on LLM outputs blocks redistribution.
- •Damien Miller questioned who could legally hold copyright on LLM-generated code, highlighting unresolved authorship and licensing issues.