EU AI Act compliance timeline for product teams.

Banned practices such as social scoring, manipulative subliminal techniques and certain biometric categorisation in public spaces are no longer allowed.

Product angle: Most product teams will not build prohibited systems, but check whether any ‘dark pattern’ or coercion features in your AI workflow fall into this bucket.

Obligations for GPAI model providers come into force, including transparency about training data and systemic-risk reporting for high-impact models.

Product angle: If you are integrating a third-party foundation model rather than training your own, this date mainly matters through your provider’s terms and any downstream obligations you inherit.

Article 50 disclosure rules for AI systems that interact with people, generate synthetic content, recognise emotions or biometric categories, or publish public-interest text begin to apply. Most high-risk AI system obligations also apply.

Product angle: This is the main UX deadline for product teams. Chatbots need first-interaction notices, generated media needs labels, deepfakes need disclosure and emotion-recognition features need exposure notices.

High-risk AI systems that are safety components of regulated products (for example certain medical, automotive or machinery use cases) must be fully compliant.

Product angle: If your AI feature is part of a regulated product, this is your final hard deadline. Use the extra year to align conformity-assessment processes with your disclosure UX.