Dynamic adversarial AI agents
Adapt the test. Preserve the evidence.
How can an AI agent adapt an authorised security assessment while keeping its actions bounded and findings reproducible?
A security assessment changes as evidence is collected. We explore agents that revise their testing priorities within an authorised scope, retain the observations behind each decision and give reviewers enough context to reproduce and assess a finding.
Research areas
Study adaptive testing priorities informed by observations and uncertainty.
Maintain explicit scope, action limits and human review boundaries.
Preserve evidence linking a suspected weakness to a reproducible assessment.
Evaluate useful coverage, false positives and testing cost in controlled environments.
Publications & related resources
Projects, publications and research materials will be added here.
Contribute to this research.
Explore student research, internships and collaboration with the lab.