AI Is Creating a New Industry: Independent AI Testing
September 19, 2026 · Keith Pasley
I keep seeing the same name pop up behind some of the more interesting stories about AI agents and cybersecurity: Irregular.
They’re a frontier AI security lab. They put advanced models into realistic cyber environments and test what they’re actually capable of doing. Their work has been used by organizations including OpenAI and Google DeepMind. (Irregular)
Why should we know who they are?
Because as AI moves from answering questions to taking actions, companies like Irregular are helping expose what these systems can actually do when given tools, access and objectives.
That feels increasingly important for anyone making enterprise AI decisions.
We spend a lot of time talking about which model is smartest.
I think we should spend more time asking:
What can this model or agent actually do once we give it access to our environment?
Irregular is one company I’ll be watching.
Irregular — Frontier AI Security
Technology doesn’t create business value. Better decisions do.