Free Store
Get started

ISO/IEC 42001:2023 · certifiable standard

ISO 42001 certifies how you manage AI. Quox gives you the evidence.

The first standard you can be certified against for how you manage AI. An accredited body audits the management system, not the model. Quox is not certified and cannot certify you: it is the evidence layer for that system.

See the management loopFor auditors
Published18 December 2023Statuscertifiable, by an accredited bodyAnnex A38 controls · 9 objectivesQuox statusnot certified, evidence layer
60 seconds: the Plan, Do, Check, Act loop, which parts are yours, and the evidence Quox keeps for the audit.Narration is AI-generated.

Start here

An auditor certifies the system around the AI, not the AI

If you are meeting this for the first time because a customer asked, or because you need AI and need to stay on the right side of compliance, this is the idea that matters most.

What the audit looks at
  • Your AI policy and who owns it
  • Roles, responsibilities and authority
  • How you find and treat AI risk
  • The controls you chose, and why
  • Records that the system runs as described
  • Internal audit, management review, corrective action
What it does not test
  • Whether the model is accurate
  • Whether any single output is correct
  • Whether the model is free of bias

A certificate says you manage AI in a controlled, repeatable, evidenced way. It is not a quality mark on the model.

7clauses to auditClauses 4 to 10, the same numbering as ISO 27001.
38Annex A controlsThe reference set you consider and justify.
9control objectivesAnnex A groups the 38 controls under A.2 to A.10.
3years, typicallyCertificate validity, with annual surveillance audits.
  1. Stage 1Documentation and readiness reviewAccredited body
  2. Stage 2Audit of whether the system is implemented and worksAccredited body
  3. CertificateTypically valid for 3 yearsIssued by the body, not ISO
  4. SurveillanceAnnual audits in betweenRecords must still exist

ISO/IEC 42001:2023 was published jointly by ISO and the IEC on 18 December 2023. ISO writes the standard and does not certify. That is done by accredited certification bodies, and the engagement is yours.

The management loop

Plan, do, check, act, around clauses 4 to 10

The standard is one closed loop. You plan the system, run it, check it works, fix what you find, and go round again. Annex A, the 38 controls, wraps the whole thing. Same shape as ISO 27001, aimed at AI.

Planclauses 4 to 7Doclause 8Checkclause 9Actclause 10AIMScontinualimprovementAnnex A · 38 controls · 9 objectives

Plan

Clauses 4 to 7Set up the system

  • Context of the organisation

    Work out what your AI is, who it touches, and what regulators, customers and staff expect of it. This sets the scope of the system the auditor will test.

    Yours

  • Leadership

    Top management owns it. There is a written AI policy, and named people with the authority to act on it.

    Where Quox fitsThe policy is yours to write. Every governed action carries an identity, so “who was responsible” has a record behind it.

    Partly Quox

  • Planning

    Identify AI risks and opportunities, assess the impact of your AI systems on people, and set objectives. This is judgement work.

    Yours

  • Support

    Resources, competence, awareness, communication, and documented information. The auditor will ask where the records are.

    Where Quox fitsThe records themselves: witnessed, hash-chained and exportable. Training your people stays yours.

    Partly Quox

Do

Clause 8Run it under control

  • Operation

    Run the AI systems the way you said you would: operational controls in place, risks treated, impact assessments carried out. This is where the system is lived, not written.

    Where Quox fitsThe policy gate acts before a governed action runs, human approval sits on the risky steps, and each leaves a receipt in the witnessed record that it fired.

    Strongest fit

Check

Clause 9Prove it works

  • Performance evaluation

    Monitor and measure, run internal audits, and hold management reviews. The auditor wants to see that you looked, not that you meant to.

    Where Quox fitsA tamper-evident operational record to monitor from: actions taken, blocked and sent to a human. Exportable as a bundle your internal auditor, and the external one, verifies offline.

    Strongest fit

Act

Clause 10Fix it, go round again

  • Improvement

    Handle nonconformities, take corrective action, and keep improving the system. A finding that is never closed is the classic audit failure.

    Where Quox fitsPolicy changes keep an append-only history, and audit-required changes need a second person. Corrective action leaves a trail. Deciding what to correct stays yours.

    Partly Quox

Annex A

38 controls in nine objectives, and how much of each is Quox

You are not expected to apply all 38. You consider each, apply what fits and justify what you leave out. Here is an honest read, objective by objective, on what a tool like Quox can contribute. Most of it is plainly your job.

A.2

Policies related to AI

Writing the AI policy is your decision. Quox enforces the rules you write at the moment of action.

Yours
A.3

Internal organisation

Roles and accountability are yours to assign. Identity on every action and a reviewer path for risky steps give them teeth.

Partial
A.4

Resources for AI systems

An inventory of which agents exist and which tools and data they can reach. Competence and budget are yours.

Partial
A.5

Assessing impacts of AI systems

Impact assessment is judgement about people and society. Quox does not do that thinking.

Yours
A.6

AI system life cycle

Records of how a system operated are the fit here. Design, verification and validation of the model remain your engineering work.

Strong on records
A.7

Data for AI systems

Data quality and provenance are yours. Policies can block personal data reaching an external model.

Yours
A.8

Information for interested parties

Evidence bundles let a customer or auditor check what happened. What you tell users is yours to write.

Partial
A.9

Use of AI systems

Responsible use and human oversight at the point of action: the policy gate, plus approval before a risky step runs.

Strong fit
A.10

Third-party and customer relationships

Supplier due diligence and customer contracts are yours. The record shows what your agents did with a vendor’s system.

Yours

Annex B is the implementation guidance for these controls. Annex C lists potential AI objectives and risk sources, and Annex D covers using the system across sectors. Read the standard on the ISO page for the control text.

The honest map

ISO 42001 certifies that you manage your AI well. Quox gives you the evidence that you did.

The kind an auditor can check without trusting you. Here is what an audit day tends to ask, and what the record answers.

Show me how this AI system actually operated last quarter.

A witnessed record of every governed agent action, hash-chained so an edit shows. Counts you can open, not a reconstruction from memory.

How do you know your controls were working, not just written down?

Blocks, denials and approvals recorded in the same chain. A control with a receipt is a control you can show fired.

Who was accountable for that action?

An identity attached to every governed action, and a human approval where the policy required one. A governed agent cannot approve its own real gate.

Did anyone change the rules, and who agreed?

An append-only policy history. Audit-required policy changes need a second person.

Can I check this without trusting you?

An exportable evidence bundle that verifies offline, with no Quox account. The auditor recomputes it rather than taking your word.

Keeping the evidence through the cycle

A certificate typically lasts three years, with an audit every year in between. Evidence you cannot produce a year later is not evidence. ISO 42001 does not set a retention period, so you decide one from your own policy, your sector and the audit cycle. Our guide walks through how to choose.

The auditor page shows what the other side of the table can verify without a Quox account. The compliance suite covers report exports, the SoA builder and the risk tools. Working from the US framework first? The NIST AI RMF page covers it, and the two fit together.

How long to keep AI agent logs

Honest maturity

What ships today, what does not, and what is yours

Quox is one layer of a management system. The certification itself is a human engagement with an accredited body. Here is the line, drawn plainly.

Ships today

  • Witnessed, hash-chained, tamper-evident record of every governed agent action (WARD and VOLT, the evidence protocols), the documented information an auditor will sample.
  • Policy gate enforced at execution, with two-person approval for audit-required policy changes and an append-only history, so the policy is enforced and not only written.
  • Human-in-the-loop approval inbox, for the human oversight your AIMS defines: a governed agent cannot self-approve a real gate.
  • Replayable traces of every governed run (Trace Explorer, AEE envelope standard).
  • Exportable evidence bundles an auditor verifies offline, with no Quox account.
  • ISO 42001 listed among the compliance suite’s framework report exports, built from your records.

Not shipped

  • A policy simulator, so you can test a policy before it goes live. Planned, not shipped.
  • A published control-level crosswalk to all 38 Annex A controls. Not published; this page maps at the level of the clauses and the nine objectives.

Still your job

  • The certification engagement itself: choosing an accredited certification body, the Stage 1 and Stage 2 audits, and closing the findings. Quox does not do this and cannot.
  • Setting the scope, writing the AI policy and naming the owners.
  • Impact assessments, risk appetite and the risk treatment decisions.
  • Internal audit and management review. Quox supplies the record they draw on, not the reviews.
  • Testing the model for accuracy and bias. Certification does not audit that either.

The path

From nothing to Stage 1, in six moves

Timelines vary with size and how much management system you already run. The order is the useful part.

  1. Start from what you already have

    If you run ISO 27001, you already share the clause structure (Annex SL). Reuse the document control, internal audit and management review you have. Many teams extend an existing system rather than build a second.

  2. Run a gap analysis

    List your AI systems, including the quiet ones, and compare what you do today with clauses 4 to 10 and the Annex A controls. The gaps become your plan.

  3. Build the AI management system

    Scope, policy, roles, risk and impact process, and the controls you will apply and justify. Two clear pages beat a binder nobody reads.

  4. Run it and collect evidence

    This is the long step, and where the audit is won. Put the gate in front of risky actions and record what the systems do, so evidence is made as you go rather than assembled the week before.

  5. Audit yourself

    Run an internal audit and a management review, fix what they find, and keep the records. Both are required clauses, not extras.

  6. Engage the certification body

    Stage 1, then Stage 2, then the certificate. That engagement is yours and the body’s. Keep the evidence through the surveillance audits that follow.

Questions people actually ask

ISO 42001: the FAQ

What is ISO/IEC 42001?

ISO/IEC 42001:2023 is the first international standard for an AI management system (AIMS). It sets requirements to establish, implement, maintain and continually improve a management system for the responsible development and use of AI. The first edition was published on 18 December 2023.

Can a company be certified against ISO 42001?

Yes. Unlike the NIST AI RMF, ISO 42001 is a certifiable standard. An accredited third-party certification body audits your organisation and, if you pass, issues the certificate. ISO itself does not certify anyone.

Does certification mean the AI model is accurate or safe?

No. Certification audits the management system: the policies, roles, risk process and controls around your AI. It does not test the model’s accuracy. A certified organisation can still ship a weak model, and the point of the system is to catch and fix that.

How long does the certificate last?

Typically three years, with annual surveillance audits in between. That is why records matter more than a one-off document set: the evidence has to exist for the whole cycle.

What are the 38 Annex A controls?

Annex A is the reference set of controls, 38 of them under nine control objectives (A.2 to A.10), covering AI policy, internal organisation, resources, impact assessment, the system life cycle, data, information for interested parties, use of AI systems and third-party relationships. You consider each, apply what fits and justify what you leave out. Annex B gives implementation guidance.

How does ISO 42001 relate to ISO 27001 and the NIST AI RMF?

ISO 42001 uses the same clause structure (Annex SL, the shared backbone of modern ISO management-system standards) as ISO 27001, so the two can be run as one management system. The NIST AI RMF is a voluntary US framework with no certificate, often used internally to manage risk, and the resulting system can then be certified against ISO 42001. Read the NIST AI RMF page.

Is Quox ISO 42001 certified, and does it make me certified?

No to both. Quox holds no ISO 42001 certificate, and no tool can make you certified: only an accredited certification body can. Quox is the evidence layer that helps you run the system and prove it, with witnessed records, pre-action policy gates, human approval and offline-verifiable evidence.

How long should I keep AI agent logs for the audit?

ISO 42001 does not set a number. Retention comes from your own policy, your sector and the audit cycle: the evidence needs to be there for the Stage 2 audit and each surveillance audit after it. Our guide to AI agent log retention walks through how to decide. How long to keep AI agent logs.

Sources

Related frameworks

  • NIST AI RMF: voluntary AI risk framework.
  • SOC 2: CPA attestation over an audit period.
  • ISO 27001: certifiable security management system.
  • HIPAA: US health data Security Rule.
  • GDPR: EU personal data law.
  • EU AI Act: EU law for AI systems.

From the blog

How long to keep AI agent logs

The retention question the standard leaves to you, across a three-year audit cycle.

Read the article

How to Prove Your AI Behaved

What an auditor can verify without taking your word for it.

Read the article

Self-hosted · free · governed

Walk into the audit with the records already made

Certification audits your AI management system, and records are what it samples. Make them as the work happens, on your own servers, and verify them offline.

See the governance layerFor auditorsThe compliance suiteTalk to us