What is at stake?
Map the agent's permissions, tools and actions against the organisation's risk tolerance.
CYBERNETIC / AI ASSURANCE
Autonomy needs
accountability.
We're building RiskStriker to help organisations understand how AI agents fail, what the consequences are, and which controls hold up.
AGENT SECURITYOPERATIONAL RESILIENCEEVIDENCE FOR RISK DECISIONS
OUR FIRST PRODUCT / IN DEVELOPMENT
RiskStriker
Put agent autonomy
to the test.
An agent can read records, change systems and send messages. Our focus is what happens when those actions go wrong, and whether the surrounding controls contain the impact.
Explore the platform ↗THE ASSURANCE LOOP
01Set the red lineCONTEXT
02Test the failureSCENARIO
03Trace the impactCONSEQUENCE
04Retest the controlEVIDENCE
THE WORK AHEAD
From a failed test
to a better decision.
RiskStriker is being developed around three practical questions.
What actually happened?
Measure the records accessed, changes made or messages sent. An attempt alone is not the outcome.
What changed after the fix?
Repeat the scenario after remediation and compare the observed impact.
INITIAL FOCUS
Consequential agents.
Public-sector context.
We're starting with the security and resilience questions facing teams developing action-taking public-sector agents. The work is at the R&D stage.
Read the evaluator overview ↗