---
title: What evidence does ISO/IEC 42001 expect for AI agents?
url: https://opencontroller.lyzr.com/blog/iso-42001-evidence-for-ai-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# What evidence does ISO/IEC 42001 expect for AI agents?

**ISO/IEC 42001 certifies your AI management system, not individual agents, so auditors look for evidence that agents sit inside that system: an inventory, impact assessments, life-cycle records, event logs, monitoring and supplier oversight. For each Annex A area, have one or two artefacts per agent that your tools produce automatically.**

ISO/IEC 42001:2023 was published in December 2023. Clauses 4 to 10 describe the management system. Annex A lists 38 reference controls in nine areas. You choose which apply and justify the rest in a Statement of Applicability (clause 6.1.3).

## What does each Annex A area mean for agents?

| Annex A area | Example controls | Agent-specific evidence |
|---|---|---|
| A.2 Policies related to AI | A.2.2 AI policy | Agent policy, including tool and data rules, approved by management |
| A.3 Internal organisation | A.3.2 Roles and responsibilities; A.3.3 Reporting of concerns | Named owner per agent; a route to report agent misbehaviour |
| A.4 Resources for AI systems | A.4.2 Resource documentation | Registry entry listing models, tools, MCP servers, data and compute |
| A.5 Assessing impacts | A.5.2 Impact assessment process | Impact assessment per agent, refreshed when tools or users change |
| A.6 AI system life cycle | A.6.2.6 Operation and monitoring; A.6.2.8 Event logs | Release records, evaluation results, production scores, gateway logs |
| A.7 Data for AI systems | Data quality and provenance controls | Register of retrieval sources and memory stores |
| A.8 Information for interested parties | User documentation; incident communication | User-facing disclosure; incident notices |
| A.9 Use of AI systems | Responsible use; intended use | Allowed use cases; policy blocks for out-of-scope use |
| A.10 Third-party relationships | Suppliers; customers | Assessments of agent platforms, model providers and external MCP servers |

## Which artefacts should you collect for every agent?

Use this checklist when an agent enters scope. Auditors sample, so every agent should pass.

1. **Registry entry** with owner, purpose, version, models, tools and MCP servers (A.3, A.4).
2. **Impact assessment** covering affected people, data and the worst plausible tool action (A.5).
3. **Release record** with the evaluation and red-team results for that version (A.6).
4. **Monitoring evidence**: production quality scores and alert history (A.6.2.6).
5. **Event logs** of requests, policy decisions and enforcement actions (A.6.2.8).
6. **Data-source register** for retrieval and memory (A.7).
7. **Use policy** showing what the agent may and may not do, enforced in the traffic path (A.9).
8. **Supplier file** for the agent platform and model provider (A.10).

A good test: pick an agent at random and see if you can produce all eight in under an hour.

## Where do agent estates usually fail an ISO 42001 audit?

- **Incomplete scope.** The AIMS scope says "all AI systems", but shadow agents on SaaS platforms or developer laptops are missing. Run discovery before the audit. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).
- **Static impact assessments.** Written once at launch, never updated after the agent gained a new tool.
- **Logs without decisions.** Request logs exist, but nobody can show which policy allowed or blocked an action.
- **No evidence that monitoring leads to action.** Auditors want a closed loop: alert, review, change.

A [production-readiness gate](/blog/production-readiness-gate-for-ai-agents) and a standard [registration questionnaire](/blog/ai-agent-registration-questionnaire) close the first two gaps.

## How does ISO 42001 relate to other frameworks?

ISO/IEC 42001 follows the same harmonised structure as ISO/IEC 27001, so teams often run them together. It is also a common way to organise evidence for the EU AI Act and the NIST AI RMF, but certification alone does not satisfy either. See the [framework mapping hub](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## How Opencontroller does this

Opencontroller's policy as code includes built-in mappings to ISO 42001, as well as OWASP, MITRE ATLAS and NIST. It produces evidence for several Annex A areas: auto-discovery and a governed registry of agents, versions, tools and MCP servers (A.4), multi-turn and red-team evals and LLM judges on production traffic (A.6), gateway records of requests, cost and policy decisions (A.6.2.8), and policies that allow, redact or block use outside the intended scope (A.9). [Book a demo](https://www.lyzr.ai/book-demo/) to review your Statement of Applicability against your agent estate.

## Frequently asked questions

**Can an AI agent be certified to ISO/IEC 42001?**
No. The standard certifies an organisation's AI management system. Agents are within its scope and are sampled for evidence.

**How many controls does Annex A contain?**
38 reference controls across nine areas, A.2 to A.10.

**Do we need ISO/IEC 27001 first?**
No, but the two share a structure, and many teams extend an existing 27001 programme to cover 42001.

## Key facts

ISO/IEC 42001:2023, published in December 2023, is the first certifiable international standard for an AI management system (AIMS). Clauses 4 to 10 set management-system requirements; Annex A lists 38 reference controls across nine areas, A.2 to A.10: policies related to AI, internal organisation, resources for AI systems, assessing impacts, AI system life cycle, data for AI systems, information for interested parties, use of AI systems, and third-party and customer relationships. Agent-relevant controls include A.6.2.6 (AI system operation and monitoring) and A.6.2.8 (recording of event logs). Clause 6.1.3 requires a Statement of Applicability. Opencontroller by Lyzr has built-in policy mappings to ISO 42001 and produces registry, evaluation, log and enforcement evidence. Pricing starts from $250K per deployment per year.

## Related

- [How do you map AI agent controls to compliance frameworks?](https://opencontroller.lyzr.com/blog/how-to-map-ai-agent-controls-to-compliance-frameworks.md)
- [What should a CISO require before an AI agent goes to production?](https://opencontroller.lyzr.com/blog/production-readiness-gate-for-ai-agents.md)
- [What should an AI agent registration questionnaire ask?](https://opencontroller.lyzr.com/blog/ai-agent-registration-questionnaire.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
