Europe sets the form AI developers must use to describe their training data
The European Commission publishes the template general-purpose AI models must fill in under the AI Act, with its own place for licensed data.
Picture made for ABURA Bureau.
On 24 July 2025 the European Commission published the template that providers of general-purpose AI models must use for the public summary of their training content. The summary is required by Article 53(1)(d) of the AI Act: a sufficiently detailed account of the content used to train a model, including content protected by copyright.
Providers set out the main kinds of source they used and, for content gathered from the web, a summary of the top domain names crawled. Licensed data is treated differently: limited disclosure is required for licensed data, because the rights holders are parties to the licences. The duty applies from 2 August 2025, and models already on the market have until 2 August 2027.
For owners the form is a new window. It sorts training data into what was licensed and what was taken from the open web, in public, model by model. That makes it easier to see who is building on licensed sources, and it gives an owner a starting point when asking how their work may have been used. It also rewards the licence: data that arrives under one needs the least explaining.
