The AI Office issues its first requests for information to more than 30 artificial intelligence providers

On 1 September 2026 the European Commission, through the AI Office, began exercising the supervisory powers set out in the AI Act, sending requests for information to more than thirty artificial intelligence providers operating worldwide. The move was made public during a briefing by Commission spokesperson Thomas Regnier: «We have sent a series of requests for information, the first within the enforcement actions launched by the Commission, to more than 30 AI companies in two main areas». The action follows two distinct tracks: on one side safety and cybersecurity for the most advanced models — focused on protection against attacks, external evaluations and market monitoring —, on the other transparency and copyright. This second strand targets those who have not published the detailed summary of the content used for training and have not taken part in the informal dialogues. Backing the initiative, European Commission Executive Vice-President Henna Virkkunen reiterated on LinkedIn her intention to use every available tool to guarantee safety and transparency across the continent.

Although this is a preliminary step and not yet the formal opening of an investigation, the measure still carries clear penalty weight. Under Article 101 of the AI Act, supplying inaccurate, incomplete or misleading answers about general-purpose AI models exposes companies to administrative fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher. The real picture of the operation, however, has several documented grey areas: the main news agencies date the mailing to 1 September, while some industry analyses place it as early as 29 August, with no clarification from the Commission as to whether the gap separates the sending from the public announcement. The executive has also not disclosed the list of recipients, and the involvement of individual providers is only partly confirmed. The spokesperson cited recent exchanges with OpenAI and Anthropic on cyber risks, while Google's inclusion rests on a report by the outlet Euractiv that has not been officially verified. Neither the response deadlines nor the exact split of the thirty-plus companies between the two tracks is known.

The AI Office's intervention puts into practice the rules applicable since 2 August 2026 on prohibited practices, transparency and general-purpose models, and comes after a number of incidents involving AI models during the summer of 2026, which the Commission has not specified. The European move also lands in a context of sharp geopolitical divergence: on 2 September 2026, at a G20 meeting in Chapel Hill, United States technology adviser Michael Kratsios argued for the adoption of the «Carolina Principles», built on the rejection of regulatory constraints specific to artificial intelligence systems.

The AI Office's debut as a supervisor highlights an obvious procedural asymmetry: the European Commission demands that providers publish the summary of their training data, yet keeps confidential which companies it has contacted and how long they have been given to reply.

— Olya