From the Law of the Horse to the Workflow: Why Institutions, Not the Model, Decide the EU AI Act’s Effectiveness: Lessons from Hungary’s First-Wave Implementation

Abstract

Abstract Whether a new technology needs its own law is an old question, and for artificial intelligence (AI), it has been posed at the wrong level of abstraction. Both the case against technology-specific law and the case for regulation by design treat the AI model as the regulated object, although a general-purpose model acquires its risk only when deployed. This article relocates the regulatory object to the deployment workflow, defined as the envelope of thresholds, escalation pathways and human checkpoints that surrounds a model in use, and reads the EU Artificial Intelligence Act as management-based regulation that works only where the workflow is specifiable and a standing supervisor examines it. Hungary’s first-wave implementation shows the second condition deciding the outcome: the same obligations are supervised by the central bank in finance and by a ministry with no examination relationship in employment, so the Act’s uniform text conceals uneven supervision within one Member State. The Digital Omnibus on AI, which postpones the high-risk obligations because standards and authorities are late, opens a window to build supervision deliberately: the priority is not better definitions but better supervision.