The European Union's AI Act, endorsed by all 27 member states and approved by Parliament in March 2024, establishes legally binding obligations on general-purpose AI model developers, including mandatory risk assessment, adversarial testing, and transparency requirements for models trained at or above 10^25 FLOPs. The United States, through Executive Order 14110 (October 2023), pursues a lighter-touch approach combining voluntary industry commitments, federal reporting requirements via Defense Production Act authority (threshold 10^26 FLOPs), and non-regulatory infrastructure (NIST frameworks), while Congress has proposed but not yet enacted licensing and registration structures. The divergence reflects fundamentally different risk tolerances: the EU regulates downstream systemic risk (bias, accidents, human rights) across the value chain, the US focuses on dual-use and national security. For financial institutions, this creates regulatory fragmentation: EU subsidiaries and partners face immediate compliance obligations by mid-2024, while US operations face executive-order-backed reporting and potential future legislative changes; the asymmetry creates material operational and compliance-cost differentiation by jurisdiction.
314 other tracked developments touching the same claims.