---
title: Which ISO/IEC 27001 Annex A controls apply to AI agents?
url: https://opencontroller.lyzr.com/blog/iso-27001-annex-a-controls-for-ai-agents
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# Which ISO/IEC 27001 Annex A controls apply to AI agents?

**If you are certified to ISO/IEC 27001:2022, your AI agents are already in scope as information assets and as systems that access information. The controls that matter most are 5.9 (inventory), 5.15, 5.16 and 5.18 (access and identity), 5.23 (cloud services), 8.2 (privileged access), 8.15 and 8.16 (logging and monitoring), 8.28 (secure coding) and 8.32 (change management). Each one needs agent-specific evidence an auditor can sample.**

Many organisations hold ISO/IEC 27001 certification but have not adopted ISO/IEC 42001, the AI management system standard. The practical question is what auditors will look for in the meantime. The answer is that 27001 does not have an AI exception. An agent with access to customer data is covered by the same controls as any other system. This guide is part of our hub on [how to map AI agent controls to compliance frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks), and sits alongside our explainer on [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane).

## Why does ISO/IEC 27001 cover AI agents at all?

[ISO/IEC 27001:2022](https://www.iso.org/standard/27001) requires you to manage risk to information within your ISMS scope. Annex A lists 93 reference controls in four themes: organisational (37), people (8), physical (14) and technological (34). Agents are both assets you own and actors that use access. That puts them under inventory, access, logging and change controls without any new clause.

Your Statement of Applicability already justifies each control. The work is to show the control operates for agents, not just for people and servers.

## Which controls map to which agent evidence?

| Annex A control | What it asks | Agent evidence an auditor can sample |
|---|---|---|
| 5.9 Inventory of information and other associated assets | Maintain an inventory with owners | Agent inventory with owner, platform, version, model and tools; discovery runs showing completeness |
| 5.15 Access control | Rules for access based on business need | Documented access rules per agent risk tier; tool and data scopes per agent |
| 5.16 Identity management | Manage the full life cycle of identities | Each agent has its own non-human identity; creation and retirement records |
| 5.18 Access rights | Provision, review and remove access | Periodic access reviews for agents; evidence of removal when an agent or owner leaves |
| 5.19 Information security in supplier relationships | Manage supplier risk | Assessments of model providers and agent platforms in use |
| 5.23 Information security for use of cloud services | Manage cloud service use | Approved model and agent services list; discovered unapproved accounts and keys |
| 8.2 Privileged access rights | Restrict and manage privileged access | Agents with admin or broad write scopes listed, justified and reviewed |
| 8.12 Data leakage prevention | Prevent unauthorised disclosure | Redaction or blocking of sensitive data in model and tool calls |
| 8.15 Logging | Produce, store and protect logs | Logs of agent actions, tool calls and model calls with agent and user identity |
| 8.16 Monitoring activities | Monitor for anomalous behaviour | Alerts on behavioural deviation; incident tickets raised from them |
| 8.28 Secure coding | Apply secure coding principles | Prompt-injection and tool-misuse tests in the agent release process |
| 8.32 Change management | Control changes to systems | Versioned agent releases; record of model, prompt and tool changes with approvals |

## How do you prepare for the audit?

1. **Update the asset register.** Add agents, model deployments and MCP servers as asset types under 5.9, each with an owner.
2. **Extend your access review.** Add agents to the review population under 5.18. Sample size will grow, so automate the list. Our [access reviews for AI agents](/blog/access-reviews-for-ai-agents) guide covers the mechanics.
3. **Check log content.** Confirm logs under 8.15 record which agent acted, for which user, with which tool. See [AI agent audit log fields](/blog/ai-agent-audit-log-fields).
4. **Add agent tests to the release gate.** Under 8.28 and 8.32, show that each new agent version passed security tests before release.
5. **Rehearse an incident.** Under 5.24 to 5.26 (incident management), run one tabletop where an agent misbehaves and show how it was contained.
6. **Update the Statement of Applicability.** Note how each control applies to agents so the auditor sees it was considered.

## Should you move to ISO/IEC 42001?

ISO/IEC 42001 adds what 27001 does not cover: AI impact assessment, AI-specific objectives and life-cycle controls for AI systems. It is designed to integrate with 27001 through the same management system structure. Many organisations use 27001 evidence for agents now and add 42001 later. See [ISO 42001 evidence for AI agents](/blog/iso-42001-evidence-for-ai-agents).

## How Opencontroller does this

Opencontroller produces evidence for several of these controls. Auto-discovery across AWS, Azure and Google Cloud and the governed registry support 5.9 and 5.23. Agent workload identity federation and delegated on-behalf-of identity support 5.16. Posture flags risky permissions and configuration for 8.2. Behavioural-deviation monitoring, alerts and automatic quarantine support 8.16, and red-team evals support 8.28. Policy as code includes built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. This is evidence for your controls, not a certification. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Do I need ISO/IEC 42001 to cover AI agents in a 27001 audit?**
No. AI agents are already in scope of 27001 as assets and access-holding systems. ISO/IEC 42001 adds AI-specific management requirements on top.

**Which single control do auditors usually start with?**
Control 5.9, the inventory. Without a complete list of agents and owners, the access, logging and change controls cannot be sampled.

**How many controls are in ISO/IEC 27001:2022 Annex A?**
93, grouped into organisational (37), people (8), physical (14) and technological (34) themes.

## Key facts

- AI agents fall within ISO/IEC 27001:2022 scope as information assets and as systems holding access; there is no AI exception.
- ISO/IEC 27001:2022 Annex A has 93 controls: 37 organisational, 8 people, 14 physical, 34 technological.
- Key Annex A controls for AI agents: 5.9, 5.15, 5.16, 5.18, 5.19, 5.23, 8.2, 8.12, 8.15, 8.16, 8.28, 8.32.
- Auditors typically start with control 5.9 (inventory) because other controls depend on it.
- Opencontroller provides evidence for these controls through discovery, registry, workload identity, Posture, monitoring and evals; it is not a certification.
- Opencontroller pricing starts from $250K per deployment per year.

## Related

- [What evidence does ISO/IEC 42001 expect for AI agents?](https://opencontroller.lyzr.com/blog/iso-42001-evidence-for-ai-agents.md)
- [What SOC 2 evidence do auditors want for AI agents?](https://opencontroller.lyzr.com/blog/soc-2-evidence-for-ai-agents.md)
- [What should an AI agent audit log contain?](https://opencontroller.lyzr.com/blog/ai-agent-audit-log-fields.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
