Security of AI-generated code
More code. Faster change. Stronger review.
How can AI help reviewers find meaningful security risks when code arrives faster than it can be examined manually?
This exploratory area investigates security analysis for AI-assisted development. A review needs to understand how a change affects trust boundaries, data flows and dependencies across a repository. The research question is how to prioritise evidence of real risk while keeping review effort manageable as the size and frequency of changes increase.
Research areas
Identify security-relevant changes across files, services and dependencies, rather than reviewing each code fragment in isolation.
Combine language-model analysis with static analysis, execution evidence and targeted tests to investigate suspected vulnerabilities.
Select the repository context an AI reviewer needs to explain a finding, including its assumptions and affected paths.
Evaluate missed vulnerabilities, false alarms, review time and computational cost under repeated code changes.
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.