When I was at CISA, we organized identity governance around four capability areas: TRUST, CRED, PRIV, and BEHAVE. They were mapped to CDM program requirements, documented in federal reference architecture, and implemented across U.S. agencies.

But I’ve always thought the word “capability areas” undersells what they actually are.

They’re dimensions. A dimensional space in which every principal — every human, every device, every service account, every agent — exists at a specific coordinate, and where the enterprise’s security posture is a function of its ability to continuously characterize that position across all four axes simultaneously.

Here’s what each dimension actually measures:

TRUST — The provenance and accountability dimension.

TRUST answers: Has this principal been properly vouched for, and does that vetting remain valid?

For a human employee, TRUST encompasses suitability determination, identity proofing, background investigation, and continuous vetting. For a non-person entity, TRUST means: who is the human sponsor accountable for this principal’s existence and behavior? TRUST is not a one-time check. It’s a continuous assertion that the chain of accountability connecting this principal to a real-world authority remains intact.

When TRUST degrades — a background investigation lapses, a sponsor leaves the organization, a device loses its attestation — every other dimension is undermined. You can have perfect credentials and perfect privilege assignment and still have a TRUST degradation. What happens when the principal itself has no background to investigate at all is a question we’ll come back to.

CRED — The proof-of-claim dimension.

CRED answers: Can this principal cryptographically prove it is who it claims to be?

Credentials are the mechanism by which identity claims become verifiable assertions. PIV cards, certificates, FIDO2 passkeys, workload identity tokens — these are all CRED artifacts. What matters about CRED is not just whether a credential was issued, but whether it was issued by a trusted authority, is currently valid, is bound to exactly one principal, and is being presented by the principal it was issued to — not a stolen copy.

CRED is the dimension where most security investment has historically been concentrated. It’s also the dimension most impacted by ephemeral, autonomous principals, more on that soon.

PRIV — The authorized scope dimension.

PRIV answers: What is this principal permitted to do — and is that grant still appropriate?

Privilege is not binary. It’s a multi-dimensional surface of entitlements, roles, access rights, and authorized actions that must be continuously compared to what a principal actually needs to perform its function. The CDM framework operationalized PRIV through Privileged Access Management (PAM) and Identity Lifecycle Management (ILM) — ensuring that privilege is granted with minimum necessary scope, reviewed regularly, and revoked promptly when no longer needed.

The critical insight about PRIV: it is not a property of the principal. It is a relationship between the principal, the resource, and the context. The same principal may be authorized to read a file and unauthorized to delete it. Authorized at 9am and unauthorized at midnight. Authorized from a managed device and unauthorized from a personal one. Hold onto that idea; it gets stranger once privilege is not solely assigned in advance.

BEHAVE — The runtime conformance dimension.

BEHAVE answers: Is this principal acting in a manner consistent with its authorized purpose?

BEHAVE is where identity governance meets detection. It monitors whether the actual behavior of a principal — the actions taken, the resources accessed, the patterns of activity — is consistent with what was expected given that principal’s TRUST level, CRED, and PRIV grant. BEHAVE is the dimension that catches the credential that was legitimately issued to a principal who has been compromised. It’s the dimension that catches the administrator who is misusing legitimate access.

The ability to evaluate behavior has been difficult. Typically is has been answered by some static information in the Identity Store or some other repository, and in CDM was primarily answered by “has this person had the proper training and certificates for the role the are assigned”. But what do you compare behavior against when the principal has no history to begin with? We’ll get there.

Why “dimensions” is the right word.

A checklist is complete when you’ve checked the boxes. A dimension is never “done” — it has a current value, and that value can change. A principal’s TRUST position can degrade. A credential can expire or be compromised. A privilege grant can become excessive as a role changes. Behavior can shift in ways that indicate compromise or misuse.

Security posture is not a static state. It’s a continuous characterization of where every principal sits across all four dimensions — and whether that position is moving in ways that require a response.

That’s the framework. Next: why agentic AI isn’t just another principal type, and why it stresses all four dimensions simultaneously in ways nothing before it has.

— Ross Foard