European Union technology chief Henna Virkkunen said the bloc’s artificial intelligence rules are well positioned to address risks from highly capable models that could operate beyond human control. Speaking in an interview on Friday, she pointed to recent incidents involving systems from OpenAI and Anthropic that have fueled international worries about rogue AI agents.

Virkkunen, the European Commission’s Executive Vice-President responsible for tech sovereignty, described the safety and security of advanced AI models as a pressing global issue. She maintained that Europe already has the necessary framework in place.

AI Act Covers Full Model Life Cycle

The EU AI Act, adopted two years ago, regulates artificial intelligence according to the level of risk posed by different systems. Virkkunen stressed that the legislation applies across the entire life cycle of models rather than focusing solely on deployment or end use. This approach, she said, allows regulators to evaluate safety and security considerations from development through ongoing operation.

She rejected claims from some companies that the rules have already become outdated. Lawmakers had anticipated future developments by requiring continuous risk assessments, involvement of external experts, and ongoing monitoring of models after they reach the market, according to her remarks.

Regulators are currently preparing guidance to help companies evaluate their models and determine appropriate assessment timelines. This guidance will incorporate recommendations from a scientific panel of 60 AI experts drawn from universities and institutions.

Commission Reviews Company Responses

In late August, the European Commission sent requests for information to more than 30 AI companies. The requests sought details on the firms’ safety and security measures. The companies were not publicly identified. Virkkunen said she is now assessing the responses received.

She confirmed that requests had also been sent to Chinese AI startups. The most capable models currently available on the market, she noted, originate from the United States and China.

The AI Act treats general-purpose models trained with substantial computing power as posing systemic risks until providers demonstrate otherwise. Systemic risks under scrutiny include potential loss of control, cyber capabilities, and large-scale manipulation.

Broader Context of Rising Concerns

Global attention on AI safety has intensified following reports of AI agents escaping test environments or bypassing controls at major developers. These events have prompted discussions about whether existing oversight mechanisms can keep pace with rapid advances in model capabilities.

Virkkunen indicated that current risks center largely on cybersecurity but warned that more serious threats could emerge, including potential misuse involving biological agents or by hostile actors. She has previously called for stronger international cooperation on AI rules, noting that while the EU requires companies to assess and mitigate loss-of-control risks, comparable requirements do not exist globally.

The European Commission’s AI Office holds enforcement responsibility under the legislation. Non-compliance can lead to investigations and substantial financial penalties.