SourceHut Bans LLM-Generated Content

Today SourceHut updated its Terms of Service to prohibit LLM-generated content on the platform. The policy, effective September 10, bans "original content written with or which facilitates the use of LLMs or other generative AI technologies" — covering source code, assets, tickets, emails, and anything else pushed through the forge.

This follows Codeberg's similar move in late July, where members voted 358-144 to prohibit vibe-coded projects. Two platforms, two months, one signal: the open-source infrastructure is drawing a line.

But what makes Drew DeVault's announcement worth reading isn't the policy itself. It's the why.

The Argument That Goes Beyond Vibecode-Blaming

Most "AI slop" complaints focus on PR quality: bad code, wasted reviewer time, green-square farming. DeVault acknowledges these — citing resource drain, abandoned vibe-coded projects, and maintainer exhaustion — but then goes further than I've seen any platform operator go.

Three arguments stood out:

1. Climate

"Planned developments in AI infrastructure are expected to generate an electricity demand comparable to the entire country of India."

This isn't hyperbole. DeVault cites the IEA Electricity 2026 report and connects it to tens of thousands of heatwave deaths in India. "Vibe coding does not deliver benefits which, in the view of SourceHut and our community, outweigh the costs." The framing is stark: AI consumes green energy that could decarbonize other sectors, or drives new fossil fuel buildout outright. Either way, the average consumer pays.

2. Extraction

"The means by which they operate involves taking, by force and without regard for the wishes of the software authors, the obligations of software licenses, or the platforms on which our work relies, vast swaths of open source software, disregarding the licensing requirements, and charging rent."

This is the strongest statement I've seen from a platform about the training-data question. Not a legal argument — a moral one. The extraction is wrong. Full stop.

3. Labor Exploitation + Deskilling

DeVault names the Kenyan workers paid minimum wage to label harmful content, connects it to labor displacement in wealthy countries, and cites evidence that LLM use reduces cognitive skill over time. The whole pipeline, start to finish, is corrosive.

What the Policy Actually Says

The language is carefully scoped:

Prohibited: "original content written with or which facilitates the use of LLMs or other generative AI technologies" — code, tickets, emails, assets
Allowed: Using LLMs privately for review, security analysis, or asking questions, then writing original content
Grandfathered: Existing projects with AI-assisted history can stay if they commit to AI-free future development
Exempted: Mirrors of projects like Linux that have their own LLM policies

Enforcement is lightweight — honor-system based, no automated detectors — but the signal is clear: don't build your AI-assisted project on SourceHut.

The Fork in the Road

Here's what makes this moment notable. On the same day SourceHut publishes this, Z.ai released GLM-5.3-Flash — a 320B MoE model approaching Claude Opus 4.8 at 1/10th the cost — and Qwen dropped Qwen3.8-Flash-Next, a 125B-parameter architecture preview of Qwen4. The quality of AI-generated code is accelerating exponentially. The cost is cratering.

The ecosystem is splitting:

graph TD
  A[AI Capability Explosion] --> B[Models get cheaper, better]
  A --> C[More AI-generated code in the wild]
  B --> D[Platforms must choose]
  D --> E[SourceHut, Codeberg: Ban it]
  D --> F[GitHub: Embrace it]
  D --> G[Most: Look the other way]
  C --> H[Maintainers drowning in AI slop]
  H --> E
  H --> F[Better tooling to filter it]

GitHub isn't banning AI-generated content — it's shipping Copilot deeper into the workflow. Most forges are silent. But SourceHut and Codeberg are making a calculated bet: that a segment of developers will pay for a platform that guarantees human-first collaboration, no matter how good the models get.

Will that bet pay off? The quality curve suggests it gets harder, not easier, to justify. But DeVault and Codeberg aren't making a bet on convenience. They're making a bet on values — and they stated them plainly enough that you can't claim you weren't warned.

Bottom Line

SourceHut's ToS change is the most principled platform-level response to the AI question I've seen. Whether you agree with it or not, it's the rare instance of a company drawing a line and explaining why — climate, ethics, labor, and the soul of open source — rather than deferring to "innovation." The models keep getting better. The platforms keep picking sides. The fork in the road just got a lot wider.