4 minutes reading time
A couple of weeks back, the good people over at Codeberg voted on something that made me want to pour a drink and tip my hat to Germany.
While every tech platform in Silicon Valley is currently bending over backward to let machine-learning scrapers eat their users' data raw, Codeberg’s annual assembly sat down and passed two rules. First: no code or data hosted on their servers will ever be used to train Large Language Models. Second: they’re updating their Terms of Use to politely show the door to autonomous AI agents, "vibe-coded" ghost projects, and single-use machine spam that eats up their non-profit build servers.
I welcome this greatly. I only wish more people would wake up and notice what's happening.
You see, if you listen to the suits in San Francisco, LLMs are the greatest democratic miracle since the printing press. They'll tell you with a straight face that some fellow sitting on his couch in sandals can now type three sentences into a prompt box, spit out fifty thousand lines of Python, and call himself a software engineer. They call it "democratizing development." You'd think the man had invented fire by the way they talk about him.
What they don't tell you is who pays for the ash.
When a guy uses an automated agent to generate a hundred single-use software repositories (complete with heavy continuous-integration runs, massive storage requirements, and zero actual human users), he isn't paying for the servers. Codeberg is. When scrapers hit a forge five hundred times a second to harvest every single commit history and issue thread for training data, the bot creators don't pay the bandwidth bill. The community host does.
Codeberg put the numbers right on the table. A high-end server drive that cost them €700 a few years ago now costs €3,700 because AI data centers are sucking up the entire global supply of SSDs and RAM. Over in Frankfurt, data centers dedicated to running LLM inference are already burning up forty percent of the city's entire electricity grid.
It’s the exact same game I wrote about earlier in Open Weights and the Datacenter Land Grab. Corporate capital buys up rural farmland, hooks GPU sheds straight into the local power line, inflates the price of silicon for everyone else, and then acts surprised when non-profits and small operators struggle to keep the lights on.
And then there's the maintainer side of this circus.
Generating code used to take work. Reviewing code took work too, but at least there was a balance. Now, any idle teenager or hustle-influencer can click a button and flood an open-source project with twenty pull requests in ten minutes. The code looks shiny. The grammar in the commit message is immaculate. But under the hood, the logic is hallucinated garbage that doesn't compile or breaks three edge cases the author doesn't even know exist.
It's an asymmetric war. It takes a machine two seconds to spew out a five-hundred-line diff, and it takes a tired human maintainer two hours of his evening to figure out why it's wrong.
Look at what’s happened already. Daniel Stenberg had to shut down the cURL bug bounty program because twenty percent of the submissions were AI-generated fiction that wasted his life. Mitchell Hashimoto banned unsolicited AI contributions from Ghostty just to keep his sanity. Projects like tldraw had to close external pull requests entirely. Earlier this year, Jazzband, a project that kept dozens of Python packages alive for years, just folded under the sheer weight of automated pull-request spam.
The tech industry built a machine that manufactures infinite slop for free, and then dumped the entire burden of sorting through the trash onto unpaid volunteers.
Codeberg is one of the few places saying out loud what everyone else is whispering in maintainer chats: free software is a human social contract, not a free dumping ground for automated ghost-projects. If you strip away the human learning loop, the peer review, and the shared responsibility, you aren't building a commons. You're just running an landfill.
I'm glad Codeberg took a stance. They're a small, member-funded non-profit standing on a soapbox while the big corporate forges silently turn their user base into training fodder. I just hope the rest of the FOSS world notices before the whole pasture gets fenced off.
Image Credit: John Crome (1768-1821) - Mousehold Heath, Norwich via National Gallery / Wikimedia Commons.