EU president warns: AI agents that 'run away' are only a foretaste

In her 2026 State of the Union address, Ursula von der Leyen called AI the foundation of both the economy and national security, but stressed that the risks must be kept in check.

She pointed to the Hugging Face incident as an example and warned that the models under development will enable hacking "at a level we never thought possible".

AI agents that "escape their environment" sound almost like teenagers running away from home, but according to von der Leyen it's only a foretaste of what's to come. In her assessment, the dangers of self-improving models are becoming "ever more evident".

"If those developing the technology are clear about this, then so should we be," she said.

Von der Leyen plans to work with Canada, the UK and other partners on evaluation, verification and safety standards for AI models.

She wants to invite the big frontier labs (the companies behind the most advanced AI models) to talks and use the EU's AI Act as what she calls "a decisive tool for putting up guardrails".

According to reports, the EU doesn't even have reliable access to the most advanced cybersecurity models from the big AI labs right now.

For you as an individual or a company, this doesn't change anything here and now: it's a political declaration of intent, not new legislation. But keep an eye out: if the AI Act is tightened with concrete requirements for the big labs, it could in the long run affect which AI tools are available in Europe, and on what terms.