Agent security
What happens when an agent encounters adversarial instructions through a tool, document or integration?
RESEARCH FOCUSUnderstand the failure.
Before it matters.
Investigating the behaviour, boundaries and failure modes of enterprise AI.
The questions
guiding our work.
Original research and testing insights will appear here as they become available. These are our current areas of focus.
Enterprise AI reliability
How do systems behave when the information they need is ambiguous, incomplete or wrong?
RESEARCH FOCUSAssurance methodology
What testing evidence is useful enough to support a real deployment decision?
RESEARCH FOCUSGovernance in practice
Do documented controls hold when an AI system is given access to operational tools?
RESEARCH FOCUSDeployment risk
How do data sources, permissions and integrations change the risk of a deployed model?
RESEARCH FOCUSChanging system behaviour
Which changes should trigger a new assessment, and how can regressions be identified?
RESEARCH FOCUSHave an assurance question?
Discuss a system, a failure mode or an area that needs independent investigation.