Trustworthy AI
Evidence Cards for High-Impact AI
Lightweight, reviewable cards that document what an AI system was tested for, what failed, and what recourse exists.
Projects
Early directions scoped for student-accessible, publishable work.
Applied cryptography
ZK-PatchAttest
Zero-knowledge attestation that a device or fleet is patched and compliant, without revealing its configuration.
Security
LLM Patch Gap
Benchmarks LLM-assisted vulnerability triage with a dangerous-error metric, measuring where models help and where they mislead.
AI accountability
AgentLedger
Hash-chained audit logs with policy and taint checks that make tool-using AI agents reviewable after the fact.
Usable security
Evidence-Cards Trust Study
A human-subjects study of whether evidence cards help people calibrate trust in, and contest, AI decisions.
Systems measurement
FedEdge-Bench
A reproducible federated-learning benchmark measuring the real overhead of privacy, security, and verification on edge devices.