Wikimedia asks a different question about AI training and answers

An AI answer can use a source in more than one way. Wikimedia wants a US court to examine those uses separately, rather than treat training a model and delivering retrieved material to a reader as one copyright question. For anyone who relies on Wikipedia, the issue reaches beyond a chatbot: it concerns the people and sources that keep shared knowledge useful.

The Foundation announced its position on 30 October. The linked court document gives the earlier procedural date: a 29 October motion seeking permission to file a proposed friend-of-the-court brief in the consolidated OpenAI copyright litigation. The proposed brief supports neither party. That request and its arguments do not, by themselves, establish court acceptance or a decision. Read the original motion and proposed brief.

Two uses, two arguments

For pretraining, Wikimedia argues that using copied material to build a general-purpose model can be transformative and is likely to qualify as fair use under the US test. Its explanation connects that position to open-source development: requiring every developer to negotiate for all training material could, it argues, favour organisations with the largest resources. This is the Foundation's case to the court, not PING declaring a universal permission to train.

For retrieval-augmented generation, or RAG, the proposed brief focuses on services that fetch source material while answering a question. Wikimedia argues that reproducing works in ways that substitute for engagement with the originals is likely not fair use, and should respect relevant licensing terms. Its concern is therefore about what happens to the copied work at a particular stage of the service. The document does not describe every possible AI use as identical.

The human source chain matters

The Foundation's public explanation makes its dependence on human knowledge explicit. Wikipedia volunteers use reliable journalism and published research as sources, while AI systems use Wikipedia. Wikimedia argues that an overly restrictive outcome could obstruct useful open-source tools, while a blanket exemption could encourage publishers to close access and weaken the material volunteers depend on. These are the Foundation's stated concerns about incentives, not measured consequences of a judgment. Foundation announcement.

Wikimedia also emphasises attribution and ShareAlike in its explanation of Wikipedia's CC BY-SA 4.0 licence. Credit connects reused work to its contributors; ShareAlike keeps adaptations available under the same licensing terms. Its position is that retrieval-based uses should honour those conditions. The Foundation separately asks automated users to respect infrastructure limits. Open access to a source and the practical burden of collecting it are different questions. Read Wikimedia's explanation.

Three questions for an answer with sources

• Can you open the underlying source and check its date and context?
• Does the answer make clear which material it retrieved and which claims it adds?
• Can you identify the people or organisation responsible for the original work?

These are PING's suggested reading questions, rather than a test of a particular chatbot or a copyright checklist. A link can help a reader inspect evidence, but the presence of a link alone does not establish whether a service has met a licence or whether its use is lawful. Keep source verification and permission as separate parts of evaluating an answer.

For a student in Nairobi, a publisher in Lagos or a research team in London, the immediate benefit is a clearer question about where an answer comes from. The filing concerns US proceedings. It is not a new rule for those readers' countries, a local legal opinion or a guarantee that a tool credits sources correctly. PING has not audited a service or checked the court's live docket for acceptance or an outcome.

The distinction to retain is specific: what a model learns during development and what a product reproduces when someone asks a question deserve separate attention. Wikimedia has put that distinction before the court; readers can also use it to inspect the source trail of an answer. Explore more source-led reporting on PING's Innovation desk.