EU AI Act shifts toward enforcement

The EU AI Act is moving from legislation into enforcement, with prohibitions on unacceptable AI practices already in force from February 2025 and transparency obligations due to become enforceable from 2 August 2026. A first draft Code of Practice published on 5 March 2026 sets out voluntary guidance for marking and labelling AI-generated content under Article 50, including machine-readable identifiers, detection methods such as watermarking and verification tools.

The governance structure is also becoming operational. The AI Office within the European Commission will coordinate supervision of general-purpose AI models and work with national authorities through a mandatory Cooperation Forum. Member States must designate independent competent authorities with powers to request information, investigate, order corrective measures and impose fines. Ireland’s planned distributed model combines a new AI Office of Ireland with oversight by existing sector regulators.

Data protection bodies are pushing to keep safeguards central, warning against removing registration duties for high-risk systems and calling for limits on the use of special-category personal data for bias detection. Voluntary early compliance is gathering support through the AI Pact, while some major technology companies have not signed. The Commission is also supporting readiness through the AI Act Service Desk, Compliance Checker and a Single Information Platform, alongside broader infrastructure plans including the Digital Networks Act and AI gigafactories.