European AI Office to Begin Enforcing Rules for General-Purpose AI Models

From 2 August, the European Commission will be able to directly enforce parts of the Artificial Intelligence Act against providers of general-purpose models. It may examine technical documentation, copyright policies, descriptions of training data and, for the most capable models, assessments of systemic risk. It will also be able to impose fines. For companies, this marks the end of the period in which the new duties could be treated mainly as an exercise in preparing internal procedures. A practical problem is the capacity of the European AI Office. Euractiv reports that the office has 145 employees, but only 34 work directly on regulation and compliance. Such a small team is expected to supervise several global companies whose models change quickly, operate in many languages and may affect millions of users. The first investigations will therefore show whether the Commission chooses a narrow set of strategically important cases or attempts broader supervision. Geopolitics will also shape enforcement. European authorities need access to models, training records and safety tests, while most suppliers are based in the United States or China. Lawmaker Brando Benifei warned that access to the most advanced models has become a tool of political pressure. “Europe's sovereignty will be judged by whether it defends its own rules,” he said. For European customers, this is not merely a dispute between the Commission and model developers. When buying services, companies will need to verify who must provide documentation, report serious incidents and respond to requests from regulators. Contracts without audit rights, without a description of model changes and without a clear allocation of responsibility may make the customer's own compliance more difficult, particularly in banking, healthcare, human resources and public administration. Large providers can spread the cost of proving safety across thousands of customers. Smaller European companies do not have the same advantage. Regulation may strengthen confidence in their products, but it can also increase the cost of entering the market. The decisive question will be whether the office offers a predictable process, applies the same requirements to domestic and foreign firms, and answers technical questions quickly. Unclear supervision would instead favour companies with large legal departments. The Commission has already consulted providers and published guidance for general-purpose models. Guidance, however, cannot replace enforcement practice. The first cases will define how detailed the supporting evidence must be, what the office considers sufficient mitigation of risk and when it will act against a model available throughout the Union. Company leaders should therefore follow not only the wording of the law, but also every early decision, and adjust procurement, risk management and liability insurance accordingly.

European AI Office to Begin Enforcing Rules for General-Purpose AI Models

The full article is available to users registered at Hard Skills