AI is moving fast and most people are falling behind. 

The Rundown AI keeps you ahead of the curve. 

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.

Sign up to start learning.

-THE WIRE  THIS WEEK IN IDENTITY

Nobody planned this week. It just happened all at once.

N°01 · LAUNCH WEEK

C1, the company formerly known as ConductorOne, is running a launch week and doing a credible Drake impression while it does. Shadow AI discovery landed Monday and it is the one worth your time, because it goes hunting for the agents, MCP servers, and plaintext tokens your teams stood up without telling anybody, on laptops as well as in the cloud. Finding an API key sitting in somebody's .env file is a very different claim than finding it in your cloud console, and it is the harder half of the problem. The C1 Vault, their agentic credential vault, landed Tuesday. CEO Alex Bovee says two more drops are coming before the week is out, so by the time you read this the lineup will have changed again. C1 →

N°02 · PLATFORM

Saviynt named its new AI identity platform Zuma, and if you know, you know. Zuma shipped Tuesday alongside the company clearing 300 million in ARR, and it breaks into three pieces. Insights for discovery, Access for runtime authorization, Governance for lifecycle and review. SAVIYNT →

Here is why this is not just a pile of press releases. C1 started Monday. Saviynt shipped Tuesday. And the Model Context Protocol finalized its 2026-07-28 specification the same day Saviynt launched, which means the layer all of this sits on top of changed underneath it inside the same 48 hours. Nobody coordinated that. That is what a market looks like when a lot of people work out the same thing at once, and what they worked out is that you cannot govern what you cannot see, and you cannot see agents with the tools you bought for people. Like I said last week, must be something in the water.

Everybody in this market is suddenly saying "identity graph." Vendors put it on slides. Analysts use it in predictions. And almost none of them will tell you what it actually is, because for most of them it's a metaphor, a nice way to say "we connected some tables." I want to give you the real version, the one I'd actually build, because once you see it you can't unsee how badly the list-based model we've all been running fails the agentic era.

Here's the short version of why this matters now. Your IAM program thinks in lists. Users in a directory. Entitlements in a catalog. Service accounts in a vault. Each list answers "who exists" and "what can they access." That worked when identities were mostly human and access was mostly static. It falls apart the moment you have to answer a question that crosses the lists, and agentic AI makes every important question a crossing question.

The free half of this gives you the concept, clearly, so you can hold your own in any room where someone says the words. The second half is the build: the model, the edges that matter, and the sequence I'd use to stand one up. That part's for members. Let me start with what it really is.

A graph is just nodes and edges. The nodes are the things: people, service accounts, AI agents, applications, roles, resources, policies. The edges are the relationships between them: this person owns that service account, that service account assumes this role, this role can read that database, this agent calls that tool, that tool holds these credentials. The list model stores the nodes and throws away the edges. The graph model treats the edges as the point.

That sounds academic until you try to answer a real question. Take a simple one. An AI agent just did something it shouldn't have. In the list world you can tell me the agent exists and what permissions it was granted. You cannot easily tell me how it got to the resource it touched, because the path runs through a role it assumed, which inherited an entitlement, which was granted by a policy somebody wrote 18 months ago for a different reason. The answer to "how did it get there" lives in the edges, and you threw the edges away.

The graph keeps them. So "how did this agent reach that data" becomes a traversal, a walk from the agent node along the edges to the resource node, and the path it returns is the answer. Not a guess. The actual chain. That is the whole value proposition in one sentence: an identity graph lets you ask reachability questions and get real answers, and reachability is exactly what governance in the agentic era is about.

The reason this stopped being optional is scale and behavior at the same time.

Scale first. Non-human identities outnumber humans by roughly 45 to 1, and 144 to 1 in cloud-native environments, per the Cloud Security Alliance's May work on the non-human identity governance vacuum, and a large share of them get created inside applications, outside your IAM program's view. You are not going to reason about a hundred thousand non-human identities and their interconnections by reading lists. The cognitive model doesn't survive contact with the numbers.

Behavior second, and this is the part that's genuinely new. A human identity is relatively stable. An AI agent interprets a goal and chooses its own path to completion. It calls tools in sequences nobody designed. It can chain access in ways the people who granted each individual permission never imagined, because each grant looked reasonable in isolation. OWASP named the principle for this in the 2026 Top 10 for Agentic Applications. They call it Least Agency, and the mitigation guidance is about constraining what an agent can decide and how far its actions can reach. You cannot constrain a blast radius you cannot compute, and you cannot compute a blast radius from lists. You compute it from a graph.

And the threat is not hypothetical while you decide. Back in May, Adversa AI disclosed two vulnerability classes called TrustFall and SymJack that turn an AI coding agent into a supply-chain delivery system through a single trusted action, and they hit the exact tools your developers run, including the one I use every day. The NSA published MCP hardening guidance the same month, specifically because the protocol gives agents a path to privileged capability that most programs never modeled. When an agent gets compromised that way, the only question that matters is the reachability question. What can it touch now, directly and through every chain. That is a graph query. In a list-based program, it's a research project you start after the incident is already in motion.

So the identity graph is the data structure that makes the agentic governance questions answerable at all. Without it, "what's the maximum reach of this agent if it's compromised" is a research project. With it, it's a query. The DBIR put it about as plainly as a report like that ever does, telling you to pay special attention to service and machine accounts because those are the ones your agentic future will lean on. Identity is the control plane for agentic AI, and a control plane you can only read as lists is really just a filing cabinet.