What to control in an AI estate, and who owns it.

01The four things to control

1. What is running, and who owns it. One row per agent, with a named person. A team name is not an owner.
2. What each agent may do. The systems it can touch, the data it can reach, and what needs a human. Written down at approval, renewed when the agent changes.
3. What each agent spends. Cost recorded as a property of every action, attributed to a team and a project, with a cap that holds before spend is committed.
4. How to stop it. Who can pause an agent, how fast, and what the audit trail says it did before the pause.

02Why it is a board question now

In 2026, IBM's Cost of a Data Breach Report found that shadow AI, meaning AI used without approval or oversight, was involved in 43 percent of breaches, up from 20 percent the year before, across 602 breached organisations. The average shadow AI incident cost $5.39 million, and more than two thirds of organisations had no governance process in place to limit it.

Deloitte's State of AI in the Enterprise 2026, from 3,235 leaders in 24 countries, found that 74 percent expect to be using AI agents by 2027, 21 percent have a mature governance model for them, and about 80 percent lack decision boundaries or an audit trail for what their agents do.

The spend side is no better covered. In its Q2 2026 Pulse Survey of 204 US C-suite leaders at companies above $1 billion in revenue, KPMG found 53 percent had agents deployed and 36 percent had direct token or usage controls in place.

03Where the control point sits

An AI control plane is the layer that gives an enterprise central visibility and control over the AI actions its teams run: which agents and models run, who uses them, what they cost, what they may do, and who approved it. The term borrows from cloud infrastructure, where a control plane manages the systems running underneath it.

What separates one control plane from another is its enforcement point: where, in the path of an action, it can record, cap or stop. A dashboard that reads logs afterwards can report an overrun. A control point in the path can hold the budget before the spend is committed, route the call to the model that should take it, and stop an action that falls outside its approval.

04Who owns what

• The CIO or CTO signs, and owns the one view of what AI costs and does.
• The CFO owns the budgets, and needs them to hold before the invoice.
• Security owns scope and audit: who can see what, and the trail that proves it.
• The transformation lead champions the pilots, and needs the approval path that gets them deployed.
• Each agent's owner is the named person accountable for it, whoever built it.
• Employees adopt it, and need to be able to describe the work they want automated without joining an IT queue.

05A first month

1. Build the inventory from the places agents leave marks: vendor contracts and card statements, service accounts and API keys with no human named on them, AI features switched on inside licensed products, and separate conversations with each team about what runs on their behalf.
2. Put every new agent on one governed path, so its row is written when it is built rather than discovered later.
3. Set caps at organisation, team and project level, with a named person who can raise them.
4. Renew each approval when an agent's model, tools or reach change.
5. Test the audit trail with one question: what ran last week, what did it cost, and who approved it. The answer should take minutes.