---
title: How do you report AI agent risk to the board?
url: https://opencontroller.lyzr.com/blog/report-ai-agent-risk-to-the-board
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# How do you report AI agent risk to the board?

**Report AI agent risk to the board on one page: how many agents you run and how many are governed, how risk is trending, what went wrong this quarter, and which decisions you need from the board. Boards want direction and exposure, not dashboards.** Keep detailed telemetry in the appendix and lead with the three or four numbers that change a decision.

## What does the board actually need to know?

Directors are not asking how your guardrails work. They are asking four questions:

1. How much AI is acting on our behalf, and do we know where it is?
2. Is our exposure going up or down?
3. Has anything gone wrong, and was it contained?
4. What do you need from us?

ISO/IEC 42001 clause 9.3 (management review) expects top management to review the AI management system's performance, including nonconformities, monitoring results and opportunities for improvement. NIST AI RMF places accountability with senior leadership under the GOVERN function. A quarterly board report is a practical way to produce evidence for both.

## What is the one-page structure?

| Section | Content | Length |
|---|---|---|
| 1. Headline | One sentence: exposure up, flat or down, and why | 1 line |
| 2. Estate | Agents discovered vs registered vs governed; by risk tier | 4 numbers |
| 3. Trend | 4-quarter trend for 3–5 key risk indicators | 1 small chart |
| 4. Incidents | Count by severity; one-line summary of any high-severity event | 3–5 lines |
| 5. Regulatory | Obligations taking effect in the next 12 months | 2–4 bullets |
| 6. Decisions | What you need: risk appetite, budget, policy approval | 1–3 bullets |

Everything else, such as per-business-unit breakdowns and individual findings, goes in an appendix the board can ignore.

## Which metrics belong on the page?

Pick metrics that are hard to game and easy to explain.

- **Coverage:** percentage of discovered agents that are registered with an owner. The gap is your shadow AI.
- **High-risk agents with complete controls:** of agents tiered high, how many have passed evals, have scoped identities and have tested kill switches.
- **Over-permissioned agents:** count of agents flagged for risky permissions or configuration, and median days to fix.
- **Incidents:** count by severity, and mean time to contain (from detection to quarantine).
- **Unattributed spend:** share of model spend that cannot be attributed to a named agent.

Avoid vanity numbers such as total prompts processed. They say nothing about risk.

## How do you present a trend without false precision?

Use four quarters and a direction arrow. If a number changed because you improved discovery, say so: coverage often drops the quarter you start scanning a new cloud account, because you find more agents. Explain that in one line rather than letting the board read it as failure.

State the conditions behind any figure. "Mean time to contain: 18 minutes (3 incidents)" is honest. "Mean time to contain: 18 minutes" invites a confidence the sample does not support.

## How should incidents be described?

For each high-severity incident, one line each on:

1. What happened, in plain words.
2. Impact: customers, data, money, regulators.
3. How it was contained and how long that took.
4. What changed afterwards.

Never omit an incident because it is embarrassing. Boards lose trust faster from surprises than from bad news.

## What decisions should you ask for?

The report is wasted if it only informs. Typical asks:

- Approve the risk appetite for autonomous actions (for example, agents may not move money above a set amount without human approval).
- Endorse a policy that unregistered agents are blocked after a grace period.
- Fund coverage for a new platform or cloud.
- Accept a named residual risk until a deadline.

## How Opencontroller does this

Opencontroller supplies several of the raw numbers. Auto-discovery across AWS, Azure and Google Cloud plus connectors to third-party agent platforms gives the discovered count; the governed registry gives registered and owned agents; Posture flags risky permissions and configuration; the LLM Gateway attributes cost per agent and per version; and rogue-agent handling records alerts, quarantines and kills. "Ask" lets you query the whole agent estate in chat when a director asks a follow-up. See the pillar, [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

Related: [metrics for AI agent governance](/blog/metrics-for-ai-agent-governance) and [running an AI agent governance committee](/blog/run-an-ai-agent-governance-committee).

## Frequently asked questions

**How often should AI agent risk go to the board?**
Quarterly is common, with an out-of-cycle note for any high-severity incident or a regulatory change that affects the risk appetite.

**Who should present AI agent risk to the board?**
Usually the CISO or chief risk officer, with the AI governance lead supporting. The presenter should own the decisions being requested.

**Should the board see individual agents?**
Only high-risk agents involved in incidents or exceptions. Otherwise show aggregates by tier and business unit.

## Key facts

A board report on AI agent risk should fit one page: estate coverage (discovered vs registered vs governed), 3-5 key risk indicators over four quarters, incidents by severity with time to contain, upcoming regulatory obligations, and decisions needed. ISO/IEC 42001 clause 9.3 covers management review. Opencontroller supplies discovery counts, registry ownership, Posture findings, per-agent cost attribution and quarantine/kill records; it starts from $250K per deployment per year.

## Related

- [Which metrics show your AI agent governance is working?](https://opencontroller.lyzr.com/blog/metrics-for-ai-agent-governance.md)
- [How do you run an AI agent governance committee?](https://opencontroller.lyzr.com/blog/run-an-ai-agent-governance-committee.md)
- [What does an AI agent governance maturity model look like?](https://opencontroller.lyzr.com/blog/ai-agent-governance-maturity-model.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
