I study how people and AI systems assess evidence and make decisions under adversarial conditions.
Changing context
Unusual behavior is not necessarily malicious. A useful security system must distinguish a threat from a legitimate change in activity, show the evidence behind its assessment, and let an analyst challenge it.
T-UEBA
At RAM Laboratories, I led research on tactical user and entity behavior analytics. The work used heterogeneous temporal graphs, calibrated predictions, and analyst feedback under resource and connectivity constraints.
Human and agent interactions
My COG-SAFE work addressed trust boundaries and accountable interactions among people, agents, and tools. I’m interested in evaluating these systems when relevant evidence is missing, misleading, or presented with unwarranted confidence.