CISO · Security architect · Enterprise architect
Know what a system exposes before extending what can access it.
AI agents, new integrations and modernization programs are all asking for access to systems whose real capabilities nobody has fully enumerated. You are expected to approve that access with evidence you do not yet have.
The questions you need answered.
- What can this system actually be made to do, beyond what the API documentation lists?
- If an agent holds these credentials, what is the full set of actions available to it?
- Where are the trust boundaries, and which calls cross them?
- Which actions need explicit authorization, and where is that enforced?
- Can we audit how a conclusion about this system was reached?
Worth knowing first.
Capability, not endpoint
An endpoint is a door. A capability is what you can do once through it, including side effects two systems away.
Analysis is not assurance
Mapping behavior tells you what a system does. It does not make the system secure, and it does not replace a security assessment.
Authorization is your decision
Evidence of what a system exposes informs who may use it. It never decides it.
Before granting new access.
Enumerate capabilities
List what the system exposes, including undocumented and indirect effects. Mark each as verified, observed, inferred or open.
Draw the boundary
Which capabilities the new principal may use, which require approval, which are denied. Enforce it in the system, not in a prompt.
Test the enforcement
Confirm that denied actions are actually refused, with repeatable tests.
Keep the record
Store the evidence and the decision together, so the approval can be reviewed later.
Where Foundry fits.
Every finding carries its evidence state: verified, observed, inferred or open. How we work.
Make systems agent-ready
A capability map of what the system lets a principal do, with tested interfaces and the limits you set.
Verify behavior
Repeatable tests for the behaviors your decision rests on, with everything untested labeled as such.
Map undocumented systems
Interfaces and dependencies the documentation does not mention, so the boundary you draw matches the system that exists.
Reading for your situation.
Discovery, verification and security assurance are three different things
What behavioral discovery establishes, what verification adds, and why neither is a security audit or a compliance result.
How to define an effective software intelligence engagement
What to pin down before anyone analyzes a system for you: the decision, the authorized target, data handling, success criteria, deliverables and limits.
What to establish before connecting an AI agent to a sensitive system
The facts, boundaries and tests to have in place before an AI agent is allowed to act on a system that matters.
Next step.
We analyze the system. You make the authorization decision with the evidence in hand.