The Anti-AI Open Source Movement Has One Thing Going For It, and Everything Else Working Against It
Linus Torvalds recently made clear that Linux is not an anti-AI project. His position is characteristically pragmatic: does the code work or not? The broader ethical, environmental, and economic questions around large language models are, for Torvalds, someone else's problem. Fine. But his statement has had an unintended side effect: it handed the anti-AI software movement something to push back against.
And push back they are.
Codeberg, the European volunteer-run code hosting platform, has updated its terms of service to prohibit projects consisting primarily of AI-generated code. It also bans cryptocurrency projects, which might seem like a non-sequitur until you notice that a significant portion of blockchain evangelists simply pivoted to LLM enthusiasm after the 2022 crypto collapse. Codeberg clearly noticed.
Meanwhile, in the BSD world, NetBSD has treated LLM-generated code as presumptively tainted since 2024. OpenBSD's position is that since LLM output cannot be copyrighted, it cannot be contributed. FreeBSD, a year on from NetBSD setting the precedent, was still thinking about it. The anti-AI contingent in operating system development is real and growing, even if it moves at BSD pace.
On the Linux side, the situation is messier. Several of the most controversial infrastructure projects in the Linux ecosystem, systemd, GNOME, Wayland, Flatpak, have received substantial backing from Red Hat, now an IBM subsidiary. IBM acquired Red Hat in 2019 and has been slowly digesting it ever since. Red Hat leadership has, by all accounts, enthusiastically embraced AI. Draw your own conclusions about what that means for code quality as the tooling shifts.
For those who want a more nonconformist Linux experience, there are options emerging. Vendefoul Wolf Linux, based on Devuan (the systemd-free Debian fork), offers several desktop environments including three KDE forks: the long-running Trinity Desktop Environment, the newer Sonic Desktop (which aims to preserve X11 compatibility ahead of Wayland-only Plasma 6.8), and the Katana Desktop Environment, a KDE 4 fork last updated in September 2023. Whether any of this coheres into a genuine movement rather than a collection of niche curiosities remains to be seen.
The browser situation is no simpler. Mozilla has spent the past year steadily embedding LLM features into Firefox. Rather than offer an AI-free edition, Mozilla's response to user complaints has been to add toggle switches. Sarah Jamie Lewis, Executive Director of the Open Privacy Research Society, has estimated that around 136,000 lines of Firefox code now exist solely to support its AI functionality. She has set up a Base Browser Project to coordinate removing it across forks.
The existing fork landscape is a patchwork of priorities. Librewolf focuses on privacy and security. Waterfox started as a 64-bit optimisation project, now leans privacy, and is explicitly AI-free. The Tor Browser and Mullvad Browser have both excised substantial amounts of Mozilla code already, making them relevant starting points for anyone building a cleaner core.
Then the Mullvad political controversy surfaced. Its co-owner donated millions to an anti-immigration populist party. Suddenly, a browser beloved by privacy advocates was entangled in exactly the kind of political argument the open source world struggles most with. Is it acceptable to build on code developed by a company whose co-owner funds politics you find repugnant? These are not hypothetical questions anymore.
And this is where the anti-AI movement runs into its own structural problem.
The Xlibre X11 server project, which Vendefoul Wolf ships, is explicitly anti-DEI and members have made public statements critical of transgender people. The Hyprland project has faced similar controversies. Mullvad accepts cryptocurrency payments, which correlates, loosely but meaningfully, with certain political orientations. Codeberg has had to publicly address harassment from far-right groups who apparently view FOSS as ideologically threatening.
The crude version of this is: the anti-AI coalition contains people who are pro-DEI and people who are anti-DEI, people who want strong privacy through cryptocurrency and people who view cryptocurrency as a far-right financial tool, people motivated by labour concerns and people motivated by libertarian anti-corporate instincts. They share a dislike of LLM-generated code. On almost everything else, they are liable to be at each other's throats.
The commercial pressure behind AI-assisted development is enormous. The companies pushing LLM tooling into every software pipeline have resources that dwarf anything the FOSS world can muster. The only realistic counter is coordinated effort across projects that share the anti-AI position.
Whether people who agree on almost nothing except one thing can actually coordinate is the question nobody in this movement seems to want to answer directly.