Autonomy
They send emails, update records and call tools without a person in the loop. Checking the output afterward is too late.
Sprawl
Teams build them in Copilot Studio, run them in Claude and deploy them on their own servers. They rarely end up in one list.
Evidence
They want to see what an agent tried to do, and what was stopped.
It calls a tool it isn’t approved for
An agent reaches for a service outside its approved group.
Enzai blocks the call
Before it runs, including on your own servers. Agents that check with Enzai first are told “Denied by governance policy.”
The agent asks for access
It gives a reason, and a person approves or rejects it in Enzai.
The decision is logged
The attempt, the request and the decision stay on the record.
Why it matters
An agent can do anything its access allows. Give it more than its task needs, and one planted instruction is enough to misuse it.1 Let it act with a person’s login, and the record names the person, not the agent.2 And if it’s part of a high-risk system, the law will expect someone who can stop it, and logs kept for at least six months.3 So govern the access.
What is AI agent governance?
AI agent governance means deciding which tools and services autonomous AI agents may use, enforcing those limits while they run, and keeping evidence of every decision. It covers what an agent does, not just what a model says.
Which agent platforms does Enzai work with?
Microsoft Copilot Studio, where Enzai discovers the agents your teams build; Claude Enterprise, where it imports Claude Code and Cowork sessions and the tools they call; and agents on your own servers. Any MCP-compatible agent can also check its approvals through the Enzai MCP server.
Where does Enzai enforce the rules?
Before each call runs, next to your agents, including on your own servers and clusters. There, you block direct outbound traffic so agent calls go through Enzai.
What happens if Enzai can’t be reached?
If Enzai is enforcing controls when it loses its connection, it blocks AI traffic by default until it reconnects. You can choose to keep applying the rules it already has instead.
Does Enzai see our prompts?
Not by default. Prompts stay private unless you choose to share metadata only, a redacted prompt or the full prompt.
How do we get started?
Connect one platform, register the agents you find, and approve their services group by group.
