Diverging AI rules complicate compliance as agents raise liability questions

AI regulation is diverging across major markets, creating gaps and uncertainty for businesses. The UK’s sector-specific approach faces criticism for leaving inconsistent coverage, while proposals for a dedicated AI authority have not translated into government plans. The EU faces a crowded digital rulebook and unclear enforcement responsibilities. Its AI Omnibus has been criticised for delaying rather than simplifying requirements, although postponing high-risk rules could allow more time for standards and guidance. The US federal approach emphasises voluntary standards and industry self-policing, while China pursues comprehensive regulation and international governance leadership.

Organisations also face a widening gap between EU AI Act obligations and practical compliance capabilities, including costs, technical barriers to watermarking free-form text, and guidance that extends beyond the legislation. Policymakers have indicated enforcement will target those disregarding obligations rather than organisations making genuine compliance efforts, but that position remains untested.

Agentic systems intensify questions about human oversight and accountability because their value depends on acting with limited intervention. Responsibility for unexpected outcomes remains unsettled, and deployers should not assume they can disclaim liability. The UNCITRAL Model Law on Automated Contracting attributes automated actions to users, although it has not been enacted in any EU Member State. Businesses should assess guardrails, incident detection and decision auditability, and establish separate, ideally deterministic, security architecture before deployment.