Blog

Score the Estate, Not the Roadmap

October 28, 2026

•

7 min read

Posted by
Tal Herman
Chief Product Officer
On this page
Getting your Trinity Audio player ready...

Earlier this month I read that NIST's proposed agent identity tenets describe the identity you give an agent. None of them describe the identities it will find.

The obvious next question is how you measure the difference.

So we built a way to.

A Readiness Test Hiding in a Design Document

NIST wrote its tenets as design guidance for agent identity systems. They also make a very good exam.

Take tenet 3, zero standing privilege. As a design goal, it asks whether your agent holds rights while idle. As a readiness test, it asks something harder: how many dormant, over-permissioned privileged accounts exist inside your applications, within the agent's reach?

The first question is about your architecture diagram. The second is about your estate.

The Agent Identity Readiness workbook asks the second kind. Twenty-one questions across all ten tenets, each scored from 0 to 3. Zero means no, or don't know. Three means yes, and you can show evidence.

Four Rules Keep the Score Honest

A self-assessment is only as useful as the room is honest. So the workbook comes with four rules.

Score the estate, not the roadmap. "We're rolling out JIT next quarter" scores what is true today.

"Don't know" is zero. If nobody in the room can answer, an agent will find the answer before you do.

Score every tenet, including the ones that aren't yours. Tenets 5 and 6 belong to your agent platform. The score still tells you where you stand.

Bring three people. IAM, an application owner and the security operations lead.

That last rule matters more than it sounds. IAM knows the directory. The application owner knows the local admin account that was "temporary" in 2019. The SOC knows which identities it actually watches.

No single person can score this alone. That's rather the point.

The Zeros Will Cluster Around the House, Not the Agent

I won't pretend to know your score. But if I had to bet on where the zeros land, I'd start with four questions.

Does every account an agent can reach, including local application accounts, have a named and current owner?

Do you know which paths into your applications bypass your policy engine entirely?

Would you detect an agent's activity moving to an identity it was never assigned?

Does your evidence show what the agent actually did inside the application, or only what it was allowed to do?

None of these questions is about the agent. They are about the house it will walk into.

And each has the same shape. The control exists. It works. It covers the identities it was told about.

An ownerless account doesn't fail your ownership review. It never appears in it.

A local login doesn't fail your policy check. It never reaches it.

An agent that switches identities doesn't trip your monitoring. It leaves the frame.

A Zero Is Not a Penalty

Enterprises don't need a perfect score before their first agent ships. That would be unrealistic, and the agents would be obsolete before anyone finished.

What they need is to know where the gaps are, which of them carry real risk, and which system owns closing each one.

That's why every tenet in the workbook names an owner: the identity provider, PAM, IGA or the SOC. Most of those systems are already in place, and they enforce the tenets well for the identities they know about. The work is making sure they aren't enforcing on top of debt they can't see.

So where you can't answer a question, score it zero.

That's not a penalty. It's your first finding.

The Board Will Ask the Wrong Question

The tenets will reach your board, probably through an audit conversation or an AI governance review. When they do, the instinct will be to ask whether your agents comply.

There is nothing to comply with. These are initial design tenets published on a NIST blog, not a standard.

The better question fits on one line: what access will our agents inherit, and who owns closing it?

We put that question, three more for management, and what good looks like in 90 days on a one-page board brief. It's light on Orchid by design, so you can forward it as is.

Count Your Zeros

The workbook takes about thirty minutes with the right three people in the room. Total it, read your band, and take the next step that goes with it.

Then count your zeros. Each one is a question an agent can answer faster than your team can.

The agents will take this test either way. It's better if you go first.

‍

Related reading

Tenet summaries are Orchid's paraphrase of NIST's Cybersecurity Insights blog, "Design Tenets for Agent Identity Management Systems" (Oct 27, 2026). The tenets are an initial proposal.