Henna Virkkunen Urges AI Companies to Strengthen Testing and Human Oversight

18:15

The European Union has warned that intensifying global competition in artificial intelligence could be undermining efforts to ensure the safety and reliability of advanced AI systems, calling on developers to maintain effective control over their technologies as their capabilities continue to grow.

Henna Virkkunen, the European Commission’s executive vice-president for technological sovereignty, security and democracy and its digital policy lead, raised the concern on Wednesday during an international AI summit in Dublin. She argued that the pressure to compete could leave companies with insufficient time and resources to test their models before deployment.

The warning comes as increasingly sophisticated AI systems introduce new challenges for developers and regulators, particularly when autonomous agents can carry out actions that are difficult for their operators to anticipate or monitor.

Growing capabilities raise new safety concerns

Virkkunen said reports of incidents involving AI agents had highlighted situations in which companies were not fully aware of what their systems were doing. She stressed that developers must retain the ability to understand and control the behaviour of their technologies.

As AI models become more capable, assessing their performance and identifying potential risks is also becoming more complicated. According to the European Commission official, this makes comprehensive testing and independent expert evaluation essential components of responsible development.

Her remarks point to a growing regulatory concern: the speed at which AI products are being developed and introduced may outpace the time required to establish whether they operate safely under a wide range of conditions.

For regulators, the challenge is not simply to encourage innovation but also to ensure that companies can demonstrate an adequate understanding of their systems before those technologies are widely used.

Brussels steps up scrutiny of AI developers

The European Commission has already begun seeking further information from major AI companies about the safeguards they have introduced. According to Virkkunen, the requests cover security measures, safety arrangements and procedures for external evaluations.

The initiative forms part of the EU’s broader effort to make AI developers more accountable for the systems they build and deploy.

Under the European Union’s Artificial Intelligence Act, the bloc has established a regulatory framework designed to address risks associated with AI applications. The legislation also provides a basis for scrutiny of issues extending beyond technical safety, including compliance with copyright rules.

Virkkunen emphasized that maintaining control over AI systems is an explicit requirement under the legislation, reinforcing the Commission’s position that technological progress must be accompanied by effective safeguards.

Balancing innovation with accountability

The EU’s concerns reflect a difficult balance facing policymakers and the technology industry. Companies are competing internationally to develop increasingly powerful models, while governments are seeking assurances that these systems will remain manageable, reliable and compliant with applicable rules.

Insufficient testing could make it harder to identify vulnerabilities before deployment, particularly in systems designed to perform tasks with limited direct human intervention. Independent assessments can help identify weaknesses that internal testing may overlook, although their effectiveness depends on access to the systems and the quality of the evaluation process.

The Commission’s requests for information signal a greater focus on how AI companies assess and manage these risks in practice, rather than relying solely on their stated commitments to safety.

As AI capabilities advance, the central question for European regulators is whether developers can sustain meaningful human oversight while continuing to innovate at the pace demanded by global competition.

Brussels is making clear that, under its regulatory approach, competitiveness cannot replace the responsibility to understand, test and control increasingly powerful AI systems.