EU proposes stricter AI safety rules

The proposal sharpens existing frameworks by defining clearer categories of high-risk applications and imposing mandatory risk assessments before deployment. It seeks to address gaps identified in early implementation phases by requiring ongoing monitoring and documentation throughout a system’s lifecycle. This ensures accountability and transparency for operators and developers alike.

Controls now include strict data governance standards to prevent bias and ensure input quality, alongside requirements for human oversight in critical decisions. Systems must support interpretability so users understand how outcomes are reached, reinforcing user confidence and regulatory compliance.

Enforcement mechanisms are strengthened with expanded authority for national regulators and standardized procedures for cross-border oversight. Non-compliant products face removal from the market and potential fines, aligning penalties with the seriousness of violations.

The changes aim to foster innovation within clear boundaries, balancing progress with protection. By setting a high bar for safety and ethics, the EU hopes to influence global norms and encourage responsible development worldwide.