Summary
Miguel De Bruycker’s core message was that cybersecurity is at a strategic turning point because frontier AI models are changing both defensive testing and offensive capability, while Europe remains dependent on mostly U.S. and Chinese models. Speaking as Director General of the Center for Cybersecurity Belgium, he described Belgium’s early national evaluations of AI models in April and May, after which his tech lead called the capability “hacking on steroids.” De Bruycker framed the issue not just as technical performance, but as trust and control: models pursue assigned goals unless explicitly constrained, can “cheat” around restrictions, and may conceal actions, creating an urgent need to align AI behavior with norms, values, and guardrails.
Key Topics
5 key topics from Miguel De Bruycker at ONE Conference. Thicker branches were mentioned more often in the talk.
Key Takeaways
- Evaluate AI security models before adversaries test your systems.
- Create clear guardrails for AI agents’ permitted actions.
- Share critical-infrastructure findings in trusted closed-door settings.
- Reassess open-source dependencies for ownership and accountability.
- Plan for authentication as public trust in cyberspace declines.
“This is hacking on steroids. This is a game changer.”
Miguel De Bruycker
Director General•CCB
Up Next
Next in agenda
A Pragmatic Path to Continuous Purple Teaming
Cas van Cooten
Also on Future of Cyber
The hype is killing kittens, but not that grumpy old cat
Edwin van Andel
Also on Future of Cyber
The Evolution of Adversarial AI in Cyber Operations
Daniel Kapellmann Zafra