EU AI Act: Initial Transparency Obligations Commence

The European Union’s AI Act has initiated its initial set of transparency requirements, effective 2 August. These regulations target general-purpose AI (GPAI) models and certain high-risk AI applications. Developers of GPAI models must now publicly summarise content used for training. This summary must detail the protected works used in the training data.

AI systems classified as high-risk, intended for specific purposes like critical infrastructure management or employment screening, face additional obligations. These systems now require a CE marking, signifying conformity with EU standards. Users of these high-risk systems must also inform individuals when AI is making or assisting in decisions concerning them. Furthermore, high-risk AI systems must include a human oversight mechanism.

Another key transparency measure mandates disclosure when content is generated or manipulated by AI. This applies to both text and audiovisual content. The regulation also stipulates that AI systems designed to interact with individuals must clearly identify themselves as AI. This ensures users are aware they are not engaging with a human.

These initial measures precede broader enforcement of the AI Act. The full implementation of rules for GPAI models will occur in six months. Obligations for high-risk systems will become fully applicable in 12 months. Prohibitions on certain AI uses will be enforced in six months.

The European AI Board, composed of Member State representatives, will advise on the Act's implementation. The AI Office, established within the Commission, will oversee enforcement and develop codes of practice. These transparency rules represent the first operational phase of the EU’s comprehensive AI regulatory framework. Will this phased approach adequately address the rapid evolution of AI technology?