Ursula von der Leyen wants to invite major US AI labs to talks in Europe and offer them European expertise on controlling AI development. In her 2026 State of the Union address, the EU Commission President made clear: the AI Act should not only be a European standard, but become an instrument for global AI security standards.
Key facts
- Von der Leyen invites frontier labs to talks and wants to cooperate with Canada, the UK and other partners on model evaluation and AI security
- Concrete risks: autonomous hacking attacks (reference to Hugging Face incident) and self-improving models that the EU currently lacks full access to
- The AI Act is positioned as a central instrument to "set guardrails" and enable Europe to shape global efforts
- To date, the EU has no reliable access to the strongest cybersecurity models from major AI labs
Hacking and autonomous agents as catalyst
Von der Leyen taps into two currently dominant narratives in the AI security debate. On one hand, she warns of immediate threats:
"Models currently being developed enable hacking at a level we never thought possible."
She refers to the Hugging Face incident. Even more critical is her perspective on autonomous AI agents that could escape their environment or inject malicious code – a scenario she describes as "just a preview" of what's coming.
Self-improving models as strategic risk
The second risk narrative concerns self-improving models – AI systems that expand their own capabilities without human guidance. Von der Leyen argues here with a kind of trust logic:
"If those who develop the technology are aware of it, then we should be too."
This is remarkable rhetoric: she assumes frontier labs know the risks – and derives from this that EU regulation is justified. This creates pressure on major labs to make their security concerns more transparent, or be accused of inaction.
Multilateral alliance instead of unilateral regulation
Von der Leyen relies not only on regulation but on international cooperation. She wants to work with Canada, the UK and other partners in three areas:
| Area | Focus |
|---|---|
| Model evaluation | Standards for assessing AI systems |
| Verification | Checking security measures |
| AI security | Joint security standards |
The invitation to frontier labs is strategic: they are not just to be regulated, but drawn into dialogue. This can be read as an offer – or as a threat that the AI Act will otherwise be enforced globally without their input.
What this means for German companies
For German AI developers and companies working with AI, the situation is tightening. The AI Act will not only remain a European standard but could become the benchmark for international business relationships – especially if major US labs agree to European standards. This means: anyone working with international partners must already expect higher compliance requirements. At the same time, an opportunity emerges: German expertise in AI security and governance could be in international demand if the EU expands its role as a security pioneer.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
