ISO/IEC 42001, published in 2023, is the international standard for an AI management system (AIMS). Like ISO 27001 for information security, it describes how an organisation governs AI systematically — policies, roles, risk assessment, controls and continual improvement. As more organisations pursue certification, a practical question comes up: how do AI agents fit into ISO 42001, and what evidence do they need? This guide gives an overview; check the standard itself and your certification body for authoritative detail.

How ISO 42001 is structured

The main clauses describe the management system: context, leadership, planning, support, operation, performance evaluation and improvement. Annex A provides a reference set of AI-specific controls — commonly described as 38 controls across nine areas — which you select based on your risk assessment and justify in a Statement of Applicability. Annex B gives implementation guidance. Certification involves an audit by an accredited body, and auditors look for operational evidence, not just documents.

Where AI agents fit

Agents are AI systems that take actions, so they bring their own risks into the AIMS: unintended actions, over-broad access, and decisions that need human oversight. Your AI agent risk assessment feeds the AIMS risk process, and the controls you choose for agents should appear in your Statement of Applicability.

Annex A areas most relevant to agents

Without reproducing the standard, these areas tend to matter most for agents:

  • Policies for AI — your AI agent policy is a natural, enforceable expression of them.
  • Internal organisation and roles — named owners for each agent and approvers for high-impact actions.
  • AI system lifecycle — how agents are designed, tested, deployed, changed and retired, including controlled rule changes.
  • Use of AI systems — keeping agents within their intended purpose and limits.
  • Third-party relationships — tool providers, model providers and connectors such as MCP servers.

Evidence auditors look for

Auditors want to see controls operating, not just described. For agents, useful evidence includes:

Control intent Evidence
Agents operate within defined limits Enforced rules and records of allowed and denied actions
Human oversight for significant actions Approval records with approver, time and reason
Controlled changes Versioned rule history with reviews and dry-run results
Monitoring and review Regular reviews of decision logs and follow-up actions
Accountability Owner per agent; decisions traceable to rules and people

An AI agent audit trail provides much of this directly.

Practical steps

  1. Add agents to your AI system inventory with owners and purposes.
  2. Run a risk assessment per agent and record chosen controls.
  3. Enforce limits at runtime and keep decision records.
  4. Version and review rule changes.
  5. Review agent activity as part of AIMS performance evaluation.

These steps also support other frameworks — see AI agent compliance.

Where Gaten fits

Gaten produces the operational evidence behind several agent controls: enforced rules, approval records, versioned change history and an exportable decision log. Certification decisions rest with your auditors. See enterprise controls.

Frequently asked questions

What is ISO 42001?

An international standard (ISO/IEC 42001:2023) that specifies requirements for an AI management system — how an organisation governs the development and use of AI.

Does ISO 42001 cover AI agents?

It covers AI systems generally. Agents fall within scope, and their action-related risks should be assessed and controlled within the management system.

Is ISO 42001 certification required?

It's voluntary, though customers and partners increasingly ask for it. Some organisations use it to structure their approach to obligations under laws such as the EU AI Act.