The Registries Are Fragmented Again. We've Solved This Before.
Extending the Identity Data Fabric to the agentic enterprise — what changes today, and why the foundation matters more than the feature.
A developer spins up an agent on a Tuesday afternoon. It gets read access to a customer database to do one job. It does the job well, so it gets a little more access. Then a little more. Six months later that developer changes teams. The agent is still running. It still holds every permission it ever accumulated. And the one person who knew why it existed is now working on something else.
Nobody offboarded the agent, because nobody owns it anymore. Its access didn’t notice the owner left.
This is the Uncontrolled Inheritance Chain, and it’s one of the fastest-growing blind spots in enterprise security. Here’s the part security leaders keep telling me: it isn’t the agents your engineers build on cloud-native platforms that worry them most. It’s the ones business users stand up in low-code tools — Copilot Studio, Agentforce — outside every control you have, invisible to every program you run.
We’ve lived this problem before
If the shape of it sounds familiar, it should.
Twenty years ago, enterprises woke up with dozens of user directories and no unified view of who anyone was. Every application had its own. Every acquisition added more. Identity was scattered across systems that didn’t talk to each other, and security teams were asked to govern a population they couldn’t see in one place.
Radiant Logic was built to solve exactly that. We didn’t rip out the directories — we federated them, correlated them, and turned fragmentation into a single, trustworthy view. That’s the Identity Data Fabric. A third of the Fortune 100 and 60% of U.S. Federal agencies run on it today.
Now look at the agentic enterprise. Every hyperscaler is building a native agent registry. None of them federates across the others. An enterprise running agents on three platforms has three registries, three partial views, and no single answer to “what exists, who owns it, and what can it reach.”
Same problem. New identity class. We’ve untangled this shape once, at the scale that matters — and we’re going to do it again, on the same vendor-agnostic foundation.
You can’t govern what you can’t correlate
Here’s what the agentic security gold rush keeps getting backwards.
An agent is not a standalone object. It’s a node in a chain. It runs back to a human owner. It inherits that human’s permissions. It spawns service principals and OAuth grants beneath it. It reaches out to tools, to data, and to other agents. The risk isn’t the agent sitting in isolation — it’s the chain, and the chain crosses every source you have.
A point tool that watches one platform sees one link. A governance product bolted onto agents governs only what it’s pointed at. Neither can reconstruct the chain, because neither was built to correlate identity data across fragmented sources.
That correlation layer is the whole job — and it’s the only place an authoritative record of an agent can actually live. The authority doesn’t come from staring at the agent. It comes from connecting the agent to everything else it touches: the human behind it, the workloads beneath it, the access it has quietly accumulated. That’s the difference between a list and an answer.
It’s also why we are not selling you another silo.
What we shipped today
We extended the Unify → Observe → Act framework to AI agents — the same framework, now covering a third identity class.
Unify. We surface the agents across the frameworks you connect into a single view, capture each one’s configuration, its approved model engine, and what it can reach — and we link every agent back to its human owner and the workloads it has picked up along the way. The relationship graph, not just the inventory.
Observe. We score risk continuously, per agent, and recalculate as privilege, ownership, configuration, and behavior change. Stale agents, orphaned agents, agents whose access has drifted well past their original purpose — surfaced and prioritized in real time, not in a quarterly review nobody finishes.
Act. We produce a remediation path on a principle we don’t bend: AI recommends, humans approve, systems enforce. We are not handing autonomous control of your environment to a model. The judgment stays with your people. The execution gets automated once they’ve made the call.
The six questions I would be asking if I were your CISO
If you’re a CISO being told to “go govern the agents” by a board that read one headline, these are the six questions underneath the mandate and our honest answers.
1. How do I see the agents I don’t even know I have? Visibility comes first. The agents you don’t know about are usually the ones living in registries and platforms you’ve never viewed side by side. We surface them across the frameworks you connect and map each to an owner and its access, so what was scattered across fragmented registries finally shows up in one place. You cannot govern a population you cannot see and most organizations can see well under a quarter of theirs today.
2. When the person who created an agent leaves, who’s accountable for what it can still reach? That’s the inheritance chain, and it’s the core of what we built. Every agent links to a human owner and the access it has accumulated. The moment that owner moves on, the orphaned agent and its drifted privilege surface instead of running quietly for another year with no one watching.
3. My agents run on platforms that don’t talk to each other. How do I get one view? The same way we unified dozens of directories: federation, not replacement. We sit across the registries and give you one source of truth, without rip-and-replace.
4. Visibility is table stakes — how does this actually reduce risk? Continuous, per-agent risk scoring tied to a prioritized remediation path. Not a dashboard that’s stale by morning — a score that moves as the agent’s posture moves, with a clear next action attached.
5. I just invested in IGA and IAM. Does this replace it or work with it? It works with it. We’re the foundation underneath, not a competing stack on top. We interoperate with your existing identity, governance, and access programs through the Shared Signals Framework, so the investment you already made keeps paying off.
6. Why Radiant Logic? Because we’ve untangled identity fragmentation once already, at the scale that matters, and the agentic version is the same problem in new clothing. The foundation a third of the Fortune 100 and most of the U.S. federal cabinet already trust for human and non-human identity now extends to agents — vendor-agnostic, no new silo, on day one.
The race is on. The entry fee is a foundation.
The agentic AI race in identity has started, and the entry fee isn’t a flashy feature that demos well and ages out in a quarter. It’s a foundation that can actually see the whole chain. Every link, across every platform, back to the human who started it.
We’ve been building that foundation for two decades. As of today, it covers agents too.
Come pressure-test it. We’ll be at Identiverse 2026, June 15–18, Booth #401, and our CEO John Pritchard opens with “The Three Identity Problem: Surviving the Chaotic Era” on June 16. Bring your hardest question.
— Cameron Matthews, CISO, Radiant Logic
