The European Commission proposed a compromise for the EU's AI law, focusing on general-purpose AI (GPAI) models and introducing codes of practice for models with systemic risks. This proposal aims to address disagreements among EU countries about regulating powerful foundation models like GPT-4. The compromise maintains a tiered approach, differentiates between GPAI models and systems, and suggests obligations for models with systemic risks, including transparency requirements and risk mitigation measures. This is part of the final negotiation phase for the landmark AI Act.