On 2 August 2025, the EU AI Act’s obligations for general-purpose AI models took effect across the Union. From that date, providers placing new models on the market must comply immediately with transparency (e.g., training-data documentation) and copyright requirements; models on the market before August 2, 2025, need to meet the same by 2 August 2027. Providers of advanced, systemic-risk models (exceeding 10^25 FLOP) face additional duties—namely, Commission notification and stringent safety and security measures.
This move updates the EU’s digital strategy to ensure clearer information on AI training processes, better enforcement of intellectual-property protections, and a security-by-design approach, thereby enhancing accountability and trust in AI across member states.
Source: European Commission