Evidence over alarm
← The AI security framework
A LEADERSHIP DISCUSSION BRIEF

What can our AI do?
How would we stop misuse?

Start with one important workflow. Name an accountable owner for each area it touches, ask for a realistic control test, and record what the test leaves uncovered.

AREAASK THE OWNEREVIDENCE TO REQUEST
01 / EmployeesWhich AI tools are people using—and what are they sharing?

Show one sensitive-data test from warning to resolution, and list the apps and devices that are outside coverage.

02 / AgentsWhat can an agent do, and who gave it permission?

Test attempts to reach an unapproved system, show the enforced block, and explain the test’s limits. Demonstrate how you revoke the agent’s access.

03 / CloudIf one account is compromised, how far can it go?

Trace a test credential from initial use to detection and revocation. Explain which connected systems remain reachable.

04 / SoftwareCan we find and fix exploitable flaws before they are used?

Take one real finding through a tested fix into the running application. Show the evidence that the flaw is resolved and how to roll back.

05 / Security operationsCan defenders understand and interrupt an incident in time?

Run a realistic exercise. Measure time to a correct decision and containment, alongside false alarms and unintended disruption.

Governance across every area

Inventory → accountable owner → policy → test evidence → documented gaps → review.

Leave with a decision.

Workflow / serviceAccountable ownerNext test + due dateKnown gap + action

A qualitative discussion guide, not a security score or a compliance assessment. Relevant controls reduce or detect risk; none guarantee prevention. Guidance: NIST AI RMF 1.0, NIST CSF 2.0, NIST SSDF 1.1, OWASP, and NCSC/CISA secure AI guidance. Sources and incident examples: Faultlines / AI security framework. Reviewed September 2026.