Overview

• Operators of advanced AI models must record safety protocols and stress-test outcomes to demonstrate mitigation of systemic risks.
• The new transparency obligations for general-purpose AI systems formally took effect across the EU on 2 August.
• Mid-July guidance from the European Commission, including a voluntary code of practice, aims to help companies navigate the GPAI requirements.
• From August 2026 for new models and August 2027 for existing ones, the EU AI Office can levy fines up to €15 million or 3 percent of global turnover for noncompliance.
• Industry coalitions are calling for a two-year moratorium while author and publisher groups warn the law does not adequately safeguard copyrighted training material.