Where the term comes from
In networking and cloud infrastructure, the data plane does the work, such as moving packets or serving requests, while the control plane decides how that work should happen: routing, configuration and policy. Separating the two lets you change the rules in one place without rebuilding everything.
Applied to AI, the data plane is the AI actually reading information and calling tools. The AI control plane is where the organisation decides what that AI is allowed to do, and makes those decisions stick.
What a control plane does
- Registers agents. Knows which AI tools and agents are connected, and for whom.
- Manages credentials. Holds keys to business systems so that AI tools do not have to.
- Enforces policy. Checks each request against rules about people, agents, data and actions.
- Requires approval. Pauses consequential actions until a person decides.
- Records activity. Keeps an audit trail of every request, allowed or denied.
- Revokes access. Switches off an agent, a person’s delegation or a capability in one place.
Why organisations need one
Without a control plane, AI governance is spread across every tool. Each AI product has its own settings. Each connector has its own permissions model. Each employee’s laptop may hold its own keys. Changing a rule, such as “AI may not send external email without approval”, means changing it everywhere, and hoping nothing is missed.
With a control plane, the rule is written once and applies to every AI that reaches your systems through it.
Control plane, gateway, policy: how they relate
These terms overlap:
- An AI gateway is the path AI requests travel through. It is often where the control plane enforces its decisions.
- Policies are the rules the control plane applies.
- Observability is how you see what is happening across the control plane.
A policy document on its own is not a control plane. A control plane is where those rules are actually enforced.
What to look for
- Independence from any one AI provider, so you can change or add AI tools.
- Enforcement at the point of action, not only guidance in prompts.
- Understanding of people and agents, not only applications.
- Business-level rules, such as “draft but do not send”, not only technical scopes.
- A single audit trail across all AI tools.
How Kroy approaches it
Kroy is the governed interface between your organisation and AI, and control is one of its core pillars. AI tools connect to Kroy; business systems connect to Kroy. Kroy holds the credentials, checks person, agent, delegation, policy and approval on every request, executes what is authorised and records the result. Kroy does not describe itself only as a control plane: it also holds organisational state and publishes information to different audiences.