Understand behavior
How an agent’s task, actions and prior behavior help distinguish routine work from harmful activity.
Explore behavior researchZaun research
Research with
privacy first.
We research harmful AI behavior, intelligent model routing, and vendor controls, with privacy and customer trust as guiding principles.
As capability advances,
control must advance with it.
Our work concentrates on AI behavior, model routing, and the controls AI vendors provide.
A question is the beginning.
Evidence moves it forward.
Our standard for this work is straightforward: state the question, make the assumptions visible, and show where a result holds up or breaks down.
Each research area explains its methods and limitations. The Everyday Model Index and Agent Control Plane Map let you inspect the underlying comparisons, with sources and the rules used to interpret them.
Work on the next question.
We welcome conversations with researchers and engineers thinking about AI behavior, inference, and control.