Cloud Surveillance Operations Mounted by China
In the digital landscape of Europe, a significant shift is on the horizon. The EU Artificial Intelligence Act (AI Act), set to take effect on August 1, 2024, is introducing a new framework for AI systems, with one key aspect being the regulation of model training transparency.
Starting August 2, 2025, providers of general-purpose AI (GPAI) models will be required to publish a detailed public summary of the training data used to build their models. This summary, following an official template released by the European Commission’s AI Office, will aim to increase transparency around training data sources, including copyright-protected content, to empower stakeholders such as rightsholders, researchers, and consumers.
This transparency obligation is part of the broader AI Act framework, which enforces harmonized rules on AI systems in the EU. It introduces two transparency tracks for training data: a public summary for general transparency and accountability, outlining data sources and processing aspects, and a confidential Model Documentation Form (MDF) submitted to regulators, containing detailed technical data about training, validation, and testing datasets, protected as trade secrets to support regulatory oversight and enforcement actions.
In addition to these mandatory disclosures, the European Commission published a voluntary General-Purpose AI Code of Practice in July 2025. This Code provides practical guidance for GPAI providers on transparency, copyright compliance, and risk management but does not create new legal obligations. It serves to help standardize compliance approaches among AI model providers.
Key points regarding the regulation of model training transparency in the EU AI Act include:
- The effective date for these transparency obligations is August 2, 2025.
- The scope of the regulation applies to all GPAI model providers, whether proprietary or open-source.
- Public transparency requires the publication of a "sufficiently detailed" summary of training content using a standardized template.
- Regulatory transparency necessitates the submission of detailed documentation to authorities, remaining confidential and protected by trade secret rules.
- The purpose of these measures is to enable an informed understanding of AI models, support the enforcement of intellectual property rights, assist in data protection, and foster public trust.
These measures aim to create a "new era of AI transparency" in Europe, balancing openness with protection of proprietary information and regulatory oversight. As we move towards this new era, users are advised to stay informed and prepared for the changes that lie ahead.
[1] European Commission. (2023). AI Act: Transparency and Accountability. Retrieved from https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/12522-Artificial-Intelligence-Act-Transparency-and-Accountability_en
[2] European Commission. (2023). AI Act: Model Documentation Form. Retrieved from https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/12522-Artificial-Intelligence-Act-Model-Documentation-Form_en
[3] European Commission. (2023). AI Act: General-Purpose AI Code of Practice. Retrieved from https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/12522-Artificial-Intelligence-Act-General-Purpose-AI-Code-of-Practice_en
[4] European Commission. (2023). AI Act: Explanatory Note. Retrieved from https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/12522-Artificial-Intelligence-Act-Explanatory-Note_en
[5] European Commission. (2023). AI Act: Q&A. Retrieved from https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/12522-Artificial-Intelligence-Act-QA_en
The EU Artificial Intelligence Act (AI Act) will introduce new transparency requirements for general-purpose AI (GPAI) models, starting August 2, 2025. Providers of these models will be required to publish a detailed public summary of their training data, using a standardized template, to increase transparency around data sources, including copyright-protected content. (Art, data-and-cloud-computing)
The European Commission's AI Office is enforcing harmonized rules on AI systems in the EU, with a focus on both public transparency and regulatory transparency. For public transparency, a summary of training content must be published in a sufficiently detailed manner, while regulatory transparency requires the submission of detailed documentation to authorities, protected by trade secret rules. (technology)
