Dublin: The Irish capital, home to the European headquarters of many of the world’s largest technology firms, has become the natural place to watch a new chapter in digital regulation unfold. From 2 August 2026 the Commission’s dedicated AI Office may formally investigate and enforce the obligations that the union’s landmark artificial intelligence law places on providers of general-purpose AI models.

The date matters because it ends a grace period. Until now, developers of the most capable systems, the general-purpose models that power chatbots and a widening range of business tools, had time to align with the rules while working alongside regulators. That cushion is gone, and the AI Office can now demand compliance and impose fines on firms that fall short.

The office wields concrete powers. It can request technical documentation, run its own evaluations of a model’s capabilities and risks, order corrective measures and, where necessary, levy penalties. In effect, Europe has created a supervisor with the authority to look inside the systems that increasingly shape online life, rather than relying on companies to police themselves.

A voluntary rulebook softens the transition. The Commission published a general-purpose AI code of practice in July 2025, and although the code is not legally binding, providers who follow it can expect regulators to presume they meet the corresponding legal duties. The arrangement rewards firms that sign up with greater predictability, while leaving the underlying obligations firmly in force.

Those duties centre on transparency and accountability. Model providers must document how their systems are trained and tested, share information with downstream developers who build products on top of them, respect copyright rules and take steps to manage systemic risks that the most powerful models can pose. The aim is to make the inner workings of AI legible to regulators and business customers alike.

Industry reaction is mixed. Some developers warn that heavy documentation demands could slow releases and burden smaller European firms competing against better-resourced rivals. Others argue that clear rules actually help, giving companies a stable framework and reassuring the businesses and public bodies that buy AI tools that a credible authority stands behind them.

Enforcement will test the office’s resources and resolve. Scrutinising fast-moving, technically complex systems requires expertise that regulators worldwide are scrambling to hire. The Commission has set out how the phase begins in its enforcement notice. For the technology firms clustered around Dublin, the era of voluntary promises on AI models is giving way to one of binding law.