🌳 🐘 The Oak jungle is heading to Gartner IAM.
All Posts
November 19, 2026
5 min read
How Oak Governs Every AI Agent Like Any Other Identity
Oak finds every agent, ties it to an accountable owner, right-sizes what it can reach and controls the tools it calls.
Table of contents
Overview

Know, Own, Right-Size, Control: How Oak Governs Every AI Agent Like Any Other Identity

Every agent known, owned, right-sized and held to the tools it should call. That’s the promise of AI agent access governance, and it starts with one idea: your AI agents are identities.

Oak finds every agent, ties it to an accountable owner, right-sizes what it can reach and controls the tools it calls. It does it on the same activity-based identity graph that governs your people, service accounts and API keys, so an agent is never a back door.

The agent nobody owns

Every enterprise has one by now. Call it the finance reporting agent: built in Copilot Studio last spring by an engineer on the data team, to pull the numbers for the monthly close.

To move fast, it was wired to a shared service account with write access to every finance table. Two other agents run on the same account, and its key never expires. Thirty people invoke the agent from Teams each month. It reads one table.

The engineer who built it left in the summer. The agent still runs. When the board asks which agents can reach financial data, and who answers for them, nobody can say. Every call it makes is logged under the service account, so nobody can say what it did either.

A large estate runs hundreds like it, across Copilot Studio, Bedrock, Agentforce, n8n and the agents engineers build in-house. They arrived faster than anyone could govern them.

“The board asked if we have visibility into the agents being run by third parties that have access into our systems, and we don’t have that either.”

IT leader, manufacturing

Oak was built so none of them runs as a back door.

Four steps, one graph

Oak governs every agent in four steps, all on the same graph that governs the rest of your identities.

1. Know: see every agent

One inventory of every agent, across the platforms you run and the ones your engineers build, each with the identity it runs on today. Shadow agents included.

2. Own: give it an owner

Every agent is tied to an accountable human and to the people who use it. When its builder leaves, the agent is re-owned, not left running on its own.

3. Right-size: cut what it doesn’t use

Agents hold far more than they use. Oak shows what each one can reach beside what it actually uses, and removes the rest, with the agent’s own activity as evidence. The reporting agent with write access to every finance table ends up with read access to the one table it needs.

4. Control: approve what it calls

Every action an agent takes is weighed against the people behind it. Routine work flows, risky calls wait for a human, and every call is on record. If something goes wrong, one click stops it.

One policy for every agent

Oak decides what each agent may do. Your agents keep running where they run today, and the decision is enforced there. No rebuilt agents, and no new platform to move them to.

In practice, a manager asks an agent in Slack to prepare a briefing. The agent asks Oak for the access the task needs. Oak checks it against the person behind the request: routine asks are approved on the spot, and unusual ones go to a person with Oak’s reasoning attached. When the task ends, so does the access, and the whole exchange is on record.

No standing access left behind, and no agent working in the dark.

Not only who has access, but who should

Most tools cover one piece of the agent problem: some see the agents, some block calls, some hold the keys. None of them knows who should have what.

Oak does, because it already governs the people and non-human identities your agents act for. Every decision is read from the activity-based identity graph. No one else has one.

One graph, every identity

Agents sit beside the people and service accounts they act for, so an agent can never reach more than the people behind it.

Fits what you run

Any agent platform, beside the IGA you already have. Agent governance is part of the Oak platform, not an add-on.

Every action on record

When an auditor asks which agents reach sensitive data, and who answers for them, the answer is ready.

Straight answers

Do we need to rebuild our agents? No. Oak works where your agents already run.

Is this a separate product for agents? No. Agents are governed in the Oak platform, on the same graph as your people.

Will Oak change access without our approval? No. Oak recommends, your team approves, and every change is explained, reversible and on record.

Start with agents. Keep going.

Agents act for people and run on service accounts. Once Oak sees one, it sees the rest. The same graph that governs your agents takes you to the humans and NHIs behind them, and towards an Autonomous Identity Governance Program.

See it, understand it, act on it, with every action on record.

Bring one agent platform. We’ll show you its agents, the shared accounts they run on and the access they never use.

Built for giants. Born complete.

‍

Discover related posts