On 24 July 2025, the European Commission adopted the Template for the public summary of training content for general-purpose Artificial Intelligence (AI) models, implementing Article 53(1)(d) of the EU AI Act, which requires providers of general-purpose AI models to publish comprehensive summaries of their training data. The Template applies to both commercial and open-source AI model providers. It mandates disclosure through a standardised three-section template, covering general information, including provider identification, model details, and training data size within broad ranges across different modalities, including text, image, audio, and video. It also mandates detailed data sources, including publicly available datasets, commercially licensed content, web-scraped data with disclosure of the top 10% of domain names crawled, user data, and synthetic data. It also focuses on data processing aspects, including measures to respect text and data mining opt-outs and remove illegal content. The Template balances transparency objectives with trade secret protection by requiring narrative descriptions rather than technical details, and summaries must be updated every six months or when material changes occur. The obligation becomes effective 2 August 2025 for new models, with existing models having until 2 August 2027 to comply. It was also highlighted that enforcement by the AI Office begins on 2 August 2026, with potential fines up to 3% of annual worldwide turnover or EUR 15 million for non-compliance.
Original source