Nazarii Tsebak

Adam and I are shaping a game-security evaluation toolkit around a shared question: how can frontier models give defenders a practical advantage? We want to test that through controlled investigations, reproducible findings and evidence that a change has improved the defence.

My focus is threat modelling and model evaluation: choosing meaningful test cases, assessing what a model can establish from the available evidence, and comparing capability, latency and cost. The aim is to find an approach that is useful within a developer’s actual constraints.

I also study frontier AI economics: lab financing, compute infrastructure and the way chips, power and capital shape access to models. That informs the technical choices we want to make, from which capabilities to evaluate to what could be feasible to deploy. Studio and lab relationships would support that work.