Why we're building Guardbase

After red-teaming many of the world's leading models and agentic systems, we became convinced of one thing: companies cannot rely on prompt filters alone to stop malicious actors from manipulating AI agents into harmful actions.

AI is moving from answering questions to taking action. Agents are reading emails, accessing systems, handling data, triggering workflows, and making decisions on behalf of businesses. As this shift accelerates, the security model has to change with it. We need to be able to trust AI agents without blindly trusting everything they do.

That is the problem Guardbase is built to solve.

Our approach is simple: don't trust the agent; verify the action. Guardbase checks what an AI agent is about to do before it happens, not after. And we do not evaluate actions in isolation. We look at the full trail of what the agent has seen, decided, and done, so companies can stop risky behavior without slowing down the business.

We are here to solve one of the hardest infrastructure problems of the AI era: how businesses can safely operate autonomous systems when attackers can try to socially engineer those systems into leaking data, misusing tools, or breaking critical workflows.

We believe AI agents will become part of the core infrastructure of every modern company. Guardbase exists to make sure they can be used safely. Starting with securing agents running on the laptop.

Before founding Guardbase, Joris worked in cybersecurity and deception technology, helping organizations trick attackers and reduce breach impact. Ron has deep experience in infrastructure, having scaled video systems for major household names in the Netherlands, including Cornerstone, KLM, and McDonald's.

Pieter previously served as SVP of Innovation at Darktrace after the acquisition of his company in 2023. Lennard built one of the largest cloud providers in the Netherlands during the 2010s.