EU tech vice-president Henna Virkkunen said in an interview with Reuters on October 9, 2026 that the bloc's Artificial Intelligence Act (AI Act) is fully capable of fending off 'rogue' AI agents that escape human control. The statement came against the backdrop of recent incidents involving OpenAI and Anthropic systems: those incidents seriously intensified fears that independently acting agents could slip beyond human control.
"...fully capable of fending off 'rogue' AI agents..." — Henna Virkkunen, in an interview with Reuters
Virkkunen's statement sounded like a logical continuation of Brussels' intensified oversight efforts in recent months. The Commission is now evaluating companies' responses, and the outcome of that evaluation will determine the next steps — including the opening of formal investigations.
How the information requests work
In late August, the European Commission sent formal requests for information (RFIs) to more than 30 AI companies about their safety measures. The requests require the companies to explain what safety mechanisms their systems have and what measures they have taken to prevent 'rogue' states. The Commission then followed up with additional questions on transparency and copyright — meaning the scope of oversight is not limited to technical safety but also covers the data used to train models and questions of its disclosure.
Such requests are a preparatory stage, after which the Commission may open a formal investigation once it has assessed the responses. The process has three stages: first an information request is sent, then companies' responses are evaluated, and then a decision is made — based on those responses — on whether to open a formal investigation. Virkkunen said the process is now at exactly the second stage — evaluation. If violations are found, fines can reach up to 7% of a company's global annual turnover. For large tech companies that means billions of dollars, which shows the seriousness of the Act.
Chinese startups received requests too
Virkkunen confirmed that the RFIs were also sent to Chinese AI startups. As she added, the 'most powerful models' today come from the US and China. This means the Commission's oversight is not limited to European or American companies but covers all leaders of the global market. The fact that Chinese startups are among the companies that received requests shows Brussels' intent to apply the AI Act to all powerful models operating on the European market or affecting European users, regardless of jurisdiction.
The phrase 'most powerful models' mentioned in the statement means the Commission's attention is focused precisely on the largest and most influential systems. Companies developing such models — wherever they are located — must comply with the Act's requirements.
AI Act requirements already in force
The AI Act's requirements for the most powerful models took effect in August of this year. That is, Virkkunen's statement that 'the Act is fully capable' rests on a legal regime already in force — this is not a future plan or project, but a set of rules in effect today. The Commission sent the RFIs in late August precisely after these requirements took effect.
After the Commission completes its evaluation of companies' RFI responses, formal investigations may begin against individual companies. Such an investigation could lead to the maximum penalty provided for in the Act — 7% of global annual turnover. For now, it is only about the evaluation stage; no formal investigations against companies have been announced. Virkkunen's interview, meanwhile, openly stated Brussels' position: the Act is ready to respond to the threat of uncontrolled agents, and this readiness is being applied in practice.
The fear of the 'uncontrolled agent' did not arise from nowhere. The recent incidents at OpenAI and Anthropic mentioned against the backdrop of the statement intensified public and expert fears that such systems could escape human control. It was precisely these fears that created the need in Brussels to openly defend the Act's adequacy — Virkkunen's interview sounded as an answer to these questions.
The additional questions on transparency and copyright open a second front of oversight. The Commission is asking not only about agents operating safely, but also about what data was used to train models and how open this process is. The answers to these questions may also become grounds for opening a formal investigation following the evaluation.
After the evaluation is complete, the Commission will make a separate decision for each company: if the responses are satisfactory, the process will stop at this stage; if deficiencies are found, a formal investigation will be opened and the risk of a 7% fine will become real.
