What is an Article 53(1)(d) training-data summary?
Public summary of training content · Article 53(1)(d), Regulation (EU) 2024/1689
The EU AI Act requires every provider of a general-purpose AI (GPAI) model placed on the EU market to "draw up and make publicly available a sufficiently detailed summary about the content used for training" — Article 53(1)(d) of Regulation (EU) 2024/1689. It is among the first legal instruments anywhere that force AI companies to disclose, publicly, what their models were trained on — applicable since 2 August 2025, ahead of California's AB 2013 (operative from 1 January 2026).
What the summary must contain
The European Commission issued the mandatory template on 24 July 2025 (C(2025) 5235 final) and formally adopted it as a Commission Communication on 5 December 2025 (C(2025) 8311 final — every version archived here). A compliant summary identifies the model and provider, states the modalities and overall size of the training data, then lists the main data sources by category. The accompanying Explanatory Notice sets the level of detail; it also directs a provider who further trains a model already on the market on additional data requiring a change to the summary to update it — at six-month intervals, or sooner if that data requires a materially significant change to the summary's content — and to record the date of each update.
The template's three sections
- General information — provider identification (including the authorised representative for providers established outside the EU, Article 54), model identification and dependencies, the date the model was placed on the EU market, and the modalities, overall size and general characteristics of the training data.
- List of data sources — per category: publicly available datasets, private third-party datasets (licensed or otherwise), data crawled or scraped by or on behalf of the provider (with crawler names, crawler behavior such as respect for robots.txt, collection periods, and a summary of the top domain names scraped), user data, synthetic data, and other sources.
- Relevant data processing aspects — the measures taken to respect reservations of rights under the text-and-data-mining exception (Article 4(3) of Directive (EU) 2019/790) and to remove illegal content from the training data.
Where to read the published summaries
The publication-status table lists every tracked model; each links its archive of captured versions — for example Meta Muse Spark (a summary that has already changed between versions) or FastwebMIIA.
Who must publish, and when
Every GPAI provider, EU-based or not, whose model is placed on the EU market — including providers of free and open-source models, because the open-source exemption in Article 53(2) covers only points (a) and (b) of Article 53(1), not the summary. A downstream entity that modifies a model so substantially that it becomes the provider must publish its own summary, limited to the data used for the modification. The obligation has applied since 2 August 2025 to models placed on the market from that date, with the summary due at the latest when the model is placed on the market; models already on the market before then have until 2 August 2027. See the deadlines page for the full timeline.
What happens if a provider doesn't comply
The Commission, acting through the AI Office, can request information (Article 91), request measures up to restricting or withdrawing the model (Article 93) and impose fines on GPAI providers of up to €15 million or 3% of total worldwide annual turnover in the preceding financial year, whichever is higher (Article 101). These supervision and enforcement powers apply from 2 August 2026.
Why an archive is needed
Providers publish these summaries on their own websites — behind changing URLs, rotating tokens, consent walls, and silent edits. There is no official registry, no version history, and no guarantee yesterday's summary still says what it said. The GPAI Ledger archives every version of every summary it can find, daily, with cryptographic proof of what was published when — so researchers, journalists, rights holders and compliance teams can cite the record rather than a link that may die.
Read the sources
Regulation (EU) 2024/1689 (EUR-Lex) · Commission Explanatory Notice & template (public summary of training content) · AIAL's quality grades
Last reviewed: 2026-08-22. This page summarizes the law for orientation; it is not legal advice.