GPAI / systemic risk

Regulatory labels from the EU AI Act: General-Purpose AI (a model >10^25 FLOPs) with a "systemic risk" subcategory (>10^25 plus meeting other criteria). These require safety assessment, transparency about training data, and incident reports. The first attempt to legally define a "frontier model." The GPAI Code of Practice (2024-25) operationalizes the requirements.