EU AI Office Becomes Enforcement Authority:
Article 75a Grants On-Site Inspections, Premises Sealing, Daily 5% Turnover Penalties

On 24 July 2026, Regulation (EU) 2026/1744 — the Digital Omnibus on AI — was published in the Official Journal of the European Union, entering into force on 27 July. While much of the public debate has focused on the new compliance deadlines and prohibitions, the Omnibus introduces a structural transformation of AI governance that has received far less attention: the AI Office transitions from a purely coordination body into a fully equipped enforcement authority with exclusive jurisdiction, investigative powers, and the ability to impose daily penalties of up to 5% of global turnover.

From Coordination to Enforcement: A Structural Shift

Before the Omnibus, the AI Office served primarily as a coordination hub — issuing guidance, promoting standardization, and facilitating cooperation among national market surveillance authorities. It had no direct enforcement powers. The Omnibus rewrites Article 75 and inserts four new articles (75a through 75d) that transform the AI Office into a central enforcement body with antitrust-level powers. These powers become exercisable from 2 August 2026.

Article 75 Rewritten: Exclusive Competence

The rewritten Article 75 confers on the AI Office exclusive competence for supervision and enforcement over two categories of AI systems:

• GPAI-based systems. AI systems based on general-purpose AI models where the model and system are developed by the same provider — or by providers forming part of the same undertaking
• VLOP/VLOSE systems. Systems that constitute or are integrated into a very large online platform or very large online search engine designated under the Digital Services Act

The key extension from "same provider" to "same undertaking" means that vertically integrated enterprise groups — where one entity develops the foundation model and another builds downstream AI applications — now face centralized oversight from Brussels rather than dispersed regulation across 27 member states. Four exceptions remain (Annex I harmonization legislation products, specific Annex III systems, law enforcement and border management, and justice administration systems), which stay under national authority jurisdiction.

Article 75a: Investigation and Inspection Powers

Article 75a equips the AI Office with all the powers of a market surveillance authority under the AI Act and Regulation (EU) 2019/1020, plus three critical enforcement tools:

As NicFab observes in its analysis, anyone familiar with EU competition law will recognize this architecture. It is, in substance, a transplant of the antitrust procedural model into the field of artificial intelligence. The AI Office can now open investigations on its own initiative or following a complaint, conduct dawn raids, and seal offices — powers that were previously unthinkable for an AI regulator.

Article 75c: Fines and Periodic Penalty Payments

When the AI Office establishes non-compliance by decision, after communicating preliminary findings and offering the possibility of a structured dialogue, the penalties in Article 99(3) to (7) apply. The most consequential provision is the periodic penalty payment regime:

• Up to 5% of average daily income or worldwide annual turnover in the preceding financial year, per day
• The Court of Justice has unlimited jurisdiction and may cancel, reduce, or increase the amount
• Proceeds accrue to the general budget of the Union
• A 5-year limitation period applies both to the exercise of powers and the enforcement of decisions

Article 75b offers operators the possibility to offer commitments, which the AI Office may make binding by decision and declare that no further grounds for action exist — with the possibility of reopening proceedings in three specified cases.

Implications for Vertically Integrated Providers

This structural transformation has the most profound implications for vertically integrated AI enterprises — companies that develop both foundation models and downstream AI systems within the same corporate group. Under the previous regime, such companies faced potential supervision by 27 different national authorities, each with its own interpretation, priorities, and enforcement capacity. Now, regulatory authority is concentrated in a single central body in Brussels.

The logic is sound: general-purpose AI models are inherently cross-border in nature, and fragmented national enforcement cannot effectively regulate them. Centralization at the AI Office level is a pragmatic response to the reality of AI markets. But as NicFab's analysis cautions, the resource question remains a vulnerability: Article 64(3) states that the AI Office shall be allocated adequate resources — but this is a programmatic provision subordinated to the EU budgetary procedure, not a binding guarantee of enforcement capacity.

For global AI enterprises, 2 August 2026 is no longer a distant date. From that day forward, the AI Office wields antitrust-level enforcement powers — inspections, premises sealing, and daily penalty payments of up to 5% of global turnover. AI compliance is no longer just a product design question. It is a legal risk management imperative that demands the same organizational attention as competition law compliance.