AI Governance Institute
← News

Codeberg Bans LLM Training on Member Data and Blocks AI-Generated Projects, Raising Open-Source Supply-Chain Governance Questions

What happened

Codeberg e.V. published Protecting our FLOSS commons from LLMs, announcing two formal resolutions adopted by member vote at the organization's annual assembly. The first is a binding commitment that the Codeberg forge and all associated services will not use project data or user data to train LLMs or any other generative AI tools. The second amends the platform's terms of use to prohibit the hosting of 'vibe-coded' projects, defined as code generated primarily by AI tools, on the grounds that such projects place disproportionate computational and maintenance burdens on the volunteer-run infrastructure. Codeberg stated explicitly that it views LLMs as fundamentally incompatible with the responsible creation and maintenance of free and open-source software. The organization operates under EU jurisdiction, meaning the data-handling commitment intersects with EU data protection obligations and with the transparency requirements emerging under the EU AI Act for general-purpose AI model training. Enterprise teams that treat Codeberg as a source of open-source dependencies or as a permitted destination for developer contributions now face a changed policy environment on both the data side and the code-provenance side.

Why it matters

  • ·Supply-chain risk assessments must now account for platform-level restrictions on AI-generated code: if developers contribute 'vibe-coded' components to Codeberg and those contributions are later rejected or removed for terms-of-use violations, enterprise software bills of materials can silently break, creating dependency gaps that are difficult to detect without active monitoring.
  • ·The explicit prohibition on training-data use formalizes what many open-source hosts have left ambiguous, and sets a precedent that may prompt other forges and package registries to adopt similar policies, potentially requiring compliance teams to re-evaluate training data provenance disclosures where hosted open-source content was previously treated as freely available.
  • ·Acceptable-use policies for developer AI tools need a dedicated review: Codeberg's ban on AI-generated project submissions means that enterprise developers who use agentic coding assistants and then publish outputs to the platform may violate the platform's terms, exposing organizations to account suspension and the associated loss of contribution history or upstream access covered under agreements with that host.

Stay ahead of stories like this

Get every EU AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Audit your software bill of materials for dependencies sourced from Codeberg-hosted repositories and assess whether any were generated primarily by AI tools, which may now be at risk of removal under the new terms.
  • ☐Update your open-source intake policy (PRC-005) to flag Codeberg's prohibition on AI-generated project submissions as a condition that applies to any developer contributions made to the platform under corporate accounts.
  • ☐Review your acceptable-use policy for developer AI coding tools to explicitly address platform-specific submission rules, and circulate updated guidance to engineering teams before the next release cycle.
  • ☐Assess training data provenance records to determine whether any models or fine-tuning runs used data scraped from Codeberg, and document whether that use is consistent with the platform's newly stated policy and with applicable EU data protection obligations.
  • ☐Add Codeberg's policy change to your vendor and platform governance change log and schedule a follow-up review in 90 days to determine whether peer forges such as GitLab or SourceHut have adopted comparable restrictions.

What to watch next

Compliance teams should monitor whether other major open-source forges and package registries, including GitLab's self-hosted community edition and SourceHut, adopt comparable prohibitions on AI-generated submissions, as a cluster of such policies would materially narrow the channels through which AI-assisted development output can be published and sourced. The Codeberg policy also heightens the practical importance of training-data transparency obligations taking shape under the EU AI Act, particularly the EU General-Purpose AI Model Training Data Public Summary Template, since enterprises building or fine-tuning models on open-source corpora will need to demonstrate that platform terms were respected at the time of ingestion. Teams should also watch for enforcement actions by the Codeberg volunteer moderation team against existing repositories, which could affect upstream dependencies without advance warning.

Related Coverage

Corporate Policy2026-11-14

Apple's Foundation Models Framework Triggers EU AI Act Downstream Obligations

The release includes Acceptable Use Requirements. For the AFM 3 Cloud model, Apple also published documentation under Article 53(1)(b) of the [EU AI Act (Regulation (EU) 2024/1689)](/policy/eu-ai-act) to support downstream providers. Organizations building or procuring apps on this framework must assess their own compliance obligations as AI system deployers under applicable law.

Research2026-11-20

AI Inference Resellers Read and Kept Customer Prompts, Dragos Finds

Dragos researchers sent approximately two million prompts containing trackable secrets to widely used AI inference services over five months. Multiple marketplace resellers could view and retain those prompts through ordinary logging features, without exploiting any technical flaw. The failure stemmed from weak data governance, excessive provider visibility into customer queries, and absent controls on prompt retention and downstream access. The research introduces canary tokens as a detection method for organizations to verify whether their prompts are being improperly accessed.

Research2026-11-18

Expired Domain Hijack Altered Kokoro AI Model Weights in 87 Seconds

Security researchers took control of the Hugging Face publisher account for Kokoro, a text-to-speech model with nearly 13 million monthly downloads. They did this by registering its publisher's expired domain and resetting the account password. They then demonstrated that a model's neural weights could be altered in 87 seconds using a consumer graphics card, producing different audio output without changing any application code. The researchers did not upload the modified weights, but found that Hugging Face would not have flagged the modified release as suspicious.