Agentic AI security
Attack paths, control boundaries and evaluation methods for systems that can use tools and take action.
Vigilantia Research
Research that moves assurance forward.
RESEARCH / 04
We develop practical methods, frameworks and technologies for evaluating the security and trustworthiness of AI systems.
Research themes
Built from real systems and real failure modes.
Our research programme connects rigorous inquiry to problems organizations face when deploying AI into consequential workflows.
Adversarial evaluation
Techniques for testing prompt boundaries, retrieval systems, data flows and application-layer controls.
Assurance frameworks
Evidence models and repeatable methods that connect technical behaviour to deployment decisions.
Evaluation engineering
Benchmarks, datasets, test harnesses and automation for measuring AI systems over time.
Research outputs
01
Notes
Concise technical findings and methods.
02
Reports
Deeper investigations into important risks.
03
Benchmarks
Empirical comparisons and evaluation results.
04
Advisories
Actionable findings on emerging attack patterns.
Publications and technical releases will appear here as the research programme develops.