data governance

The European AI Office held a specialized convocation of the Scientific Panel on frontier AI safety to finalize systemic-risk evaluation metrics for models exceeding the $10^25 FLOPs training threshold and those demonstrating emerging autonomous multi-step reasoning capabilities.

The AP emphasized requirements for deployers of automated decision-making and semi-automated generative systems in the Netherlands. Deployers must verify that “human-in-the-loop” oversight is operational rather than nominal. Rubber-stamping model outputs without verifiable review interfaces constitutes an immediate breach of Article 22 GDPR and triggers preliminary enforcement under national market surveillance powers.