AI for security. Security for AI.

Research

Binary code representation

Learning representations that connect the compiled forms of a function to what it does, so analysts can find known code across processors, compilers and optimization settings.

Explainable vulnerability analysis

Moving vulnerability assessment from a score to code-level evidence an analyst can inspect, through learned detection, rule generation and patch screening.

Malware beyond known families

Comparing unfamiliar malware with the available evidence instead of forcing it into a known class, using open-set recognition, functional decomposition and authorship analysis.

Firmware and supply-chain security

Recovering the software components and capabilities of a firmware image from the image itself, without relying on the supplier.

Avionics intrusion detection

Resource-aware, adaptive and unsupervised detection for aircraft data buses such as MIL-STD-1553, built on attack analysis and controlled simulation.

Publications