What is the training content summary?
The training content summary is a public document that providers of general-purpose AI models must publish under Article 53(1)(d) of the EU AI Act, describing the content used to train the model. Since 24 July 2025 it follows a mandatory AI Office template, covering data sources and types at a level detailed enough to let rights-holders and regulators understand what went in — without requiring dataset-level disclosure of trade secrets.
Who must publish one
- Providers of GPAI models placed on the EU market — including open-weight releases;
- Fine-tuners: an entity that significantly modifies a GPAI model publishes a summary covering the modification’s training content only.
Practical notes
Provenance is far cheaper to record than to reconstruct: keep source registers as you assemble fine-tuning sets. The summary interacts with the provider’s copyright policy and TDM opt-out compliance — misalignment between the two documents is an obvious audit target. Related: TDM exception and opt-out.