PrymeIntelligence

Agent certification·Regulation

ISO/IEC 42001 certification: what it proves about your AI agents

What an ISO/IEC 42001 certificate proves to a customer or supervisor about AI agents, what it leaves unproven, and how to close the gap with evidence.

A certificate under ISO/IEC 42001 proves that an organisation runs an AI management system an accredited auditor found to meet the standard, within a stated scope. It shows AI is governed: owned, risk-assessed, controlled, monitored and improved. It does not prove that any particular agent performs correctly on your cases, or that it should be trusted with a particular action.

What does the certificate attest to, in plain words?

ISO/IEC 42001:2023 sets requirements for an AI management system. To be certified, an organisation must show it has understood its context and the AI it develops, provides or uses; set an AI policy and objectives; assigned roles and responsibilities; assessed AI risks and the impacts of AI systems on people; selected and implemented controls; monitored and measured performance; run internal audits and management reviews; and corrected what it found. Our explainer on what ISO/IEC 42001 is covers the structure.

The certificate says that, at audit, this system existed and operated across the scope, based on what the auditor sampled. Surveillance audits through the certificate's life check it keeps doing so. The process behind it is covered in what to expect from a certification audit.

Who relies on it, and for what?

Three audiences read the certificate differently.

Customers buying agents use it to shortlist vendors. It tells them the vendor governs AI deliberately, rather than shipping models and hoping.

Supervisors see it as evidence that a firm's AI governance has structure. It helps answer questions about accountability and risk management, though it is not a regulatory approval.

Boards use it to confirm management has put a system in place, and to hold it to the improvement cycle the standard demands.

None of them should read it as a statement about a specific agent's behaviour.

What does it prove about agents that act, and what not?

Here is the line, set out plainly.

It proves: - Agents in scope are known, owned and listed. - Their risks and impacts were assessed using a defined method. - Controls were chosen for those risks, with reasons for any excluded. - Performance is monitored and reviewed by management. - Problems lead to corrective action.

It does not prove: - That a given agent answers correctly on your cases, refuses what it should, or escalates when it should. - That the agent you are buying sits inside the certified scope. - That the agent is safe at the depth you want to grant it in your operation. - That the organisation complies with the EU AI Act, data protection law or sector rules. - That anything is still true after the agent's model, brief or sources changed since the last audit.

This gap matters most for agents because their risk lives in actions. A management system can be excellent while one capability of one agent, say releasing payments to new payees, is unsafe in your setting. Our answer to whether agent certification is the same as ISO certification draws the distinction in brief. The longer argument is in red-teaming finds the edges; certification proves the middle: you need evidence across the ordinary run of your cases, not only assurance that someone is managing risk.

How do you close the gap?

Use the certificate for what it proves and add evidence for what it does not.

Requirement Control Evidence it leaves
Vendor governs AI Check certificate validity, issuing body's accreditation and scope Copy of certificate and scope statement
Your agent is in scope Confirm the product and functions you use are named in the scope Scope review note
Exclusions understood Review which annex controls the vendor excluded and why Statement of applicability extract
Agent performs on your cases Certify each agent by replaying real cases from your operation Per-agent results against a floor
Authority fits the evidence Grant each agent its depth based on your results, not the vendor's certificate Depth record per agent
Evidence stays current Re-test on change and on a fixed cycle; lapse authority when overdue Certificate expiry and re-test records

Your own organisation can also certify under ISO/IEC 42001 as a deployer. That is a different certificate from the vendor's and proves your management system, not theirs. If you already hold ISO/IEC 27001, see ISO/IEC 42001 vs ISO/IEC 27001 for how to run them together.

What will a supervisor or auditor ask about the certificate?

  • Whose certificate is it, yours or your vendor's, and what does its scope include?
  • Which agents you use fall inside that scope?
  • How did you test those agents on your own cases?
  • What authority have you given each agent, and on what evidence?
  • When does each piece of evidence expire?

What does the certificate not cover?

It is not a product certification, a regulatory licence or a legal opinion. It does not cover information security in depth, which is the job of ISO/IEC 27001. It does not certify that a vendor's agents are fit for your purpose. For what certifying an agent itself involves, see our guide to what it means to certify an agent.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Where Pryme fits

Pryme treats a management-system certificate as one input and certifies the agent itself. Attest replays the agent against real cases from the customer's operation and scores them across seven gates, from grounded answers to stable behaviour. The result sets the agent's depth, and the certificate lapses after 90 days whatever happens to the management-system certificate.

Questions people ask

Is an ISO/IEC 42001 certificate enough for a bank to approve an agent vendor?

It is a strong input to vendor due diligence, but banks also need security assurance, operational resilience evidence and their own testing of the agent on their cases.

Does an ISO/IEC 42001 certificate expire?

It runs for a fixed cycle with surveillance audits in between, and lapses without recertification. The agent evidence behind it may go stale much sooner, which is why agent-level certificates should run on a shorter clock.

Can a small vendor be certified?

Yes. The standard scales to the organisation's size and AI use. A small vendor with a clear scope and working records can be certified.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.