OpenBSD swats down an ext4 port after questions about AI provenance and copyright

A functioning filesystem — with strings attached

On March 17 Thomas de Grivel posted an ext4 implementation to the openbsd-tech mailing list that, it has been reported, provides full read and write access and even passes e2fsck. It does not implement journaling, and the submission included several copyright assertions but said little about how the code was produced. It has been reported that de Grivel later acknowledged parts of the work were generated by a large language model — a revelation that set off a fierce debate over provenance, licensing, and risk.

Licence panic and a blunt response

Concerns centered on whether the submission was a derivative of the Linux ext4 implementation and thus subject to the GPL. It has been reported that the LLM in question was almost certainly trained on Linux ext4 code and documentation — not exactly a calming thought for a project that prizes clean licensing. Theo de Raadt made the stakes plain: reimplementations for interoperability are legal, sure, but the current copyright system offers no clear way for non‑human outputs to grant the permissions OpenBSD needs. His verdict was blunt: "the chances of us accepting such new code with such a suspicious Copyright situation is zero."

Bigger than one patch

This is more than a squabble over one tree. It has been reported that de Grivel initially refused to retract his copyright claims on the machine‑generated portions, then later said he would remove LLM‑produced code and leave only what he wrote himself. Who believes that now? Doubt lingers. The episode crystallizes a larger problem: an increasing number of contributors treating LLMs like code factories and submitting output without clear provenance. Who will maintain that code? Who owns it? And what happens when a legal test finally answers those questions — possibly not in the open‑source community's favor?

A warning shot for open source

OpenBSD’s hard line may look brusque, but it’s a practical defense: without clear copyright, redistribution and combination — the very lifeblood of open source — become legal landmines. The emotional core here is trust. When tools can spit out thousands of lines overnight, we lose the thread connecting author to code. That matters. It matters for security, for maintenance, and for the ethos of projects that have long lived by explicit, human‑made licenses. Welcome to the age of code you can’t be sure you own.

Sources: lwn.net, Hacker News