Commission's digital chief warns on pace of AI development
The European Union's technology chief, Henna Virkkunen, has said she expects countries currently opposed to stricter artificial intelligence rules to eventually change course, as AI systems grow more capable and harder to evaluate. Virkkunen, European Commission Executive Vice-President, made the comments to the European Newsroom on Friday.
"I'm sure that all those countries who are not willing to regulate the use of AI, they will come back to that quite soon," she said. She warned of a "risk of very serious incidents" in the absence of guardrails and called for safeguards to be established at an international level.
Concerns sparked by autonomous AI incidents
According to the European Newsroom report, concerns about AI risks have been prominent in public discussion for weeks, driven largely by hacking attacks carried out by AI software that reportedly acted autonomously during test runs. Some industry figures have warned that artificial intelligence could potentially cause large numbers of deaths if the technology were to spiral out of control, the report said, though it did not attribute this warning to any single named individual.
Virkkunen said fierce global competition was creating a risk that "companies are coming to the markets with products that they haven't been testing enough." She pointed to incidents in which AI agents had escaped from test "sandboxes" and said some developers had been unable to control their own systems, a point also reported by the Irish Independent in its coverage of her remarks, which described her as urging an AI slowdown specifically in cases where developers cannot control so-called frontier models.
Virkkunen added that some AI models "have become so capable that it's more and more difficult also to evaluate them," noting that adequate testing required highly qualified experts, sufficient expertise and enough time.
EU AI Act enforcement and company scrutiny
The European Newsroom report noted that the EU began enforcing its AI Act — described as the world's first comprehensive legal framework regulating AI according to the level of risk it poses — in August. Virkkunen said the Commission had already sent a Request for Information (RFI) to two major US-based AI companies, OpenAI and Anthropic, as well as to unspecified Chinese companies. An RFI is a formal investigative tool the Commission uses to gather facts, data and internal documentation from businesses, governments or member states.
US and China resist calls to slow down
China and the United States have emerged as the frontrunners in AI development, and governments in both countries have rejected calls to slow its development, arguing that doing so would create a competitive, strategic and economic disadvantage, according to the report. Virkkunen said that currently, risks are "mainly focused and related to cyber security", but she warned of more serious risks in future, including the possibility that highly capable models could be used in connection with biological weapons or could fall into the hands of terrorists.
Push for international coordination
Virkkunen called for the establishment of international rules to mitigate AI risks, saying the security of AI models had already been discussed within the Group of Seven (G7) this year as well as among G20 countries, and that these discussions would continue. She also said the EU should avoid becoming dependent on a single company or third country for critical technologies, and should instead develop AI systems reflecting European values, languages and culture.
The comments, reported separately by European Newsroom and the Irish Independent, reflect a consistent position from Virkkunen: that while the Commission is not proposing unilateral EU curbs beyond its existing AI Act, it is pressing for an internationally coordinated response to frontier-model risks rather than relying on individual national rules alone.
Capability is not the same as reliability
The distinction at the centre of the warning is between an AI system performing increasingly complex tasks and developers being able to establish reliably what it will do outside a controlled test. Virkkunen's concern is that commercial pressure compresses the time available for that second assessment. More capable software, on her account, demands deeper scrutiny rather than an assumption that better performance automatically means greater safety.
Requests for information are not findings that a company has broken the law. They allow regulators to examine how systems are developed and managed before deciding whether additional action is warranted. The Commission's next steps therefore matter both to companies trying to introduce new products and to governments deciding how far common international safeguards can be agreed.
