Scale · founder · 7 min read

A Major Code Host Just Banned Vibe-Coded Projects. Here's What Founders Should Take From It

Codeberg's members voted 70% to ban primarily-AI-generated projects. It won't hit most founders directly — but the reasons behind it should shape how you ship.

Published July 22, 2026 ·

open source copyright governance vibe coding risk founders

Codeberg — a community-run, non-profit alternative to GitHub — just did something no major code host had done before. Its members voted to ban projects that are mostly generated by AI. The vote wasn’t close: around 70% in favor, roughly 358 to 144 with a handful of abstentions. The platform’s terms of use now prohibit repositories that consist largely of unreviewed output from tools like Claude Code and OpenAI Codex.

If you build with AI, the headline sounds alarming. It mostly isn’t. But the reasoning behind the ban is worth your attention, because it’s the first time a platform has turned the “vibe-coded slop” complaint into an actual rule — and the logic they used is the same logic that could eventually reach app stores, marketplaces, and the tools you rely on.

What the ban actually covers

Read the fine print and the scope is narrower than the headline. Codeberg didn’t ban AI-assisted development. Using Claude Code, Cursor, or Copilot to help you write and review code is still fine there. What’s banned is shipping a project that is primarily machine-generated and dumped onto the platform as a “finished” open-source repo without meaningful human review.

The community had a name for the thing they were voting against: “slopcoded” projects. Think low-effort repos spun up in an afternoon, published with a README the author never read, and abandoned. Codeberg is small, volunteer-funded, and runs on donated infrastructure. A flood of these projects costs them storage, moderation time, and credibility — and, as they argued, creates legal exposure they can’t afford.

So the practical line is: AI as a tool, reviewed by a human who understands and stands behind the result, is welcome. AI as the author of a project nobody has actually looked at is not.

The real reason: nobody knows who owns the code

The part founders should sit with is why the members voted this way. The loudest argument wasn’t quality or spam — it was copyright.

Code generated by a large language model has an unclear legal status. It’s not obviously owned by you, it’s not obviously in the public domain, and in some cases it may reproduce fragments of the licensed code the model was trained on. When you publish that under an open-source license like MIT or Apache, you’re making a promise — “I own this and I grant you these rights” — that you may not actually be in a position to make. Codeberg’s volunteers decided they didn’t want to host that liability at scale.

This is not a fringe concern. It’s the same issue lawyers have been circling for two years, now expressed as a hosting policy rather than a hypothetical. The provenance of AI-generated code — who wrote which line, and what rights come attached — is moving from “interesting question” to “thing platforms make decisions about.”

Does this affect your startup? Probably not directly

Let’s be honest about the blast radius. Codeberg is a principled, developer-heavy niche. If you’re a non-technical founder building a SaaS product, you’re almost certainly hosting your code on GitHub or GitLab, deploying on Vercel or Replit, and you’re not publishing your app as an open-source project for strangers to reuse. None of those platforms has announced anything like this, and there’s no sign the big commercial hosts are about to. Your private repo full of AI-assisted code is not in anyone’s crosshairs.

So if you came here worried you’d wake up to find your project deleted — you can relax. This is a small platform making a values-driven call about public, open-source repos, not a shift in how the tools you pay for treat your private work.

What you should take from it anyway

The signal matters more than the event. Two things are worth internalizing.

First, “a human reviewed this” is becoming the dividing line. Codeberg didn’t ban AI — it banned unreviewed AI. That distinction is going to show up everywhere: in platform rules, in enterprise procurement, in due diligence when you raise or sell. The founders who stay on the right side of it are the ones who can say, honestly, that a person who understands the code looked at what shipped. If your entire codebase is a black box even to you, that’s a growing liability regardless of where you host it.

Second, the copyright question isn’t going away — it’s getting operationalized. For most founders the practical risk of AI-generated code isn’t a lawsuit; it’s a licensing mess that surfaces at the worst moment, usually during an acquisition or a funding diligence process. You don’t need to panic about it, but you should know where your code came from, keep your dependencies and licenses clean, and avoid publishing AI-generated work under open-source licenses you can’t actually back. If you’re relying on open-source components, keep a record of what you used and under what terms.

The bottom line

Codeberg’s ban is a small platform drawing a line, not the start of an industry crackdown on the tools you use. You can keep vibe coding, and you can keep hosting your private product wherever you already do. But the reason behind the vote is the durable part: the world is starting to treat “primarily AI-generated, never reviewed by a human” as a distinct category — one with real copyright baggage and diminishing goodwill. Build so that a person who understands your product has always looked at what ships, keep your provenance and licenses tidy, and this entire trend becomes someone else’s problem instead of yours.