Skip to content

ISO 27001, SOC 2, the DPDP Act and manual VAPT. Book a Free Scoping Call

ISO 42001

ISO/IEC 42001 for AI Startups: Inventory, Roles and Impact Assessment

ISO/IEC 42001:2023 sets requirements for an AI management system: the policy, roles, risk and impact assessment, and lifecycle controls an organisation uses to develop, provide or use AI responsibly. For an AI startup the work starts with an inventory of every AI system you build, sell or rely on, your role for each one, and an impact assessment of how each could affect people. It follows the same management system structure as ISO/IEC 27001, so the two can share one system. A certificate is not compliance with the EU AI Act. SecureRoot helps startups build the system and prepare for audit; an accredited certification body issues the certificate.

Request an ISO implementation proposalSee ISO pricing

About this guide

Written by SecureRoot Risk Advisory LLP.

Scope

What an AI management system covers

ISO describes an AI management system as the policies, objectives and processes an organisation sets up for the responsible development, provision or use of AI systems. The current edition is ISO/IEC 42001:2023, published in December 2023, and ISO describes it as applicable to organisations of any size that develop, provide or use AI-based products or services.

For a startup, that usually means:

  • An AI policy that says what the company will and will not do with AI, approved by the founders.
  • Named roles for AI governance, risk ownership and human oversight.
  • A risk assessment and an impact assessment for each AI system in scope.
  • Controls across the AI lifecycle: data for AI, development and testing, deployment, monitoring of behaviour and performance, and retirement.
  • Transparency to users and customers about what the AI does and its limits.
  • Governance of third-party models, APIs and AI tools that the product or the team relies on.
  • Internal audit and management review, the same as any ISO management system.

You choose the scope. One product and the team that builds it is a common and defensible first boundary; the certificate names it, so customers will read exactly what is covered.

Start here

Building the AI system inventory

Every later step depends on knowing which AI systems exist. Startups routinely find more than they expected, because AI sits in features, vendor tools and internal workflows nobody catalogued.

  1. 01

    List what you build

    Models you train or fine-tune, features built on a third-party model API, retrieval pipelines, agents, scoring or recommendation logic. Record the purpose, the users, and the data each one uses.

  2. 02

    List what you rely on

    Third-party AI services inside the product, and AI tools staff use in daily work, such as coding assistants, meeting transcription or support drafting. Record what data each one can see.

  3. 03

    Record the supply chain

    For each system, note the model provider, version or update cadence, hosting location and contract terms on data use. Changes upstream can change your risk without any change in your code.

  4. 04

    Give each entry an owner

    An inventory without owners goes stale within a release or two. Tie updates to your change process so a new AI feature cannot ship without an entry.

Roles

Your role for each system, and who owns what

Two kinds of role matter. The first is your organisation's role for each AI system. Under the EU AI Act, a provider is broadly the party that develops an AI system, or has it developed, and places it on the market or puts it into service under its own name; a deployer is the party that uses an AI system under its own authority. A startup that sells an AI feature is often a provider of it and a deployer of the AI tools it uses internally.

The second is people inside the company. A small team does not need a committee, but it does need names against these responsibilities:

  • An accountable executive, usually a founder, who approves the AI policy and accepts residual risk.
  • A person who runs the AI management system day to day and keeps the inventory, assessments and records current.
  • Owners for each AI system who are responsible for its risk and impact assessment, monitoring and changes.
  • People with authority to review, override or switch off a system's output, with that authority written down.
  • Someone who reviews third-party AI providers before they are adopted.

Impact assessment

Assessing the impact of each AI system

An AI risk assessment looks at risk to your organisation. An AI impact assessment looks at what the system could do to individuals, groups and society. ISO/IEC 42005:2025 gives guidance on AI system impact assessment across the lifecycle, alongside ISO/IEC 42001.

  1. 01

    Describe the system and its context

    What it does, who uses it, who is affected by its output, and what decisions it informs or makes.

  2. 02

    Identify foreseeable harms

    Unfair or inaccurate outcomes for some groups, privacy intrusion, unsafe output, over-reliance by users, and misuse beyond the intended purpose.

  3. 03

    Record the controls and the oversight

    Data quality and provenance checks, evaluation before release, monitoring of behaviour and performance after release, user transparency, and who can review or override the output.

  4. 04

    Decide and record the residual risk

    The accountable owner accepts what remains or asks for more treatment, against acceptance criteria set in advance.

  5. 05

    Set a review trigger

    Repeat the assessment when the model, its data or its purpose changes materially, not only before an audit. A stale assessment gives little assurance.

ISO/IEC 27001

How ISO 42001 relates to ISO/IEC 27001

ISO/IEC 42001 follows the same management system structure as ISO/IEC 27001, so context, leadership, risk process, internal audit, management review and corrective action work the same way. ISO/IEC 27001 manages information security risk; ISO/IEC 42001 adds the AI-specific governance: inventory, impact assessment, data for AI, lifecycle, transparency and third-party AI providers.

  • If you already hold ISO/IEC 27001, add the AI-specific layer to the system you run rather than building a second one.
  • If customers ask for both, build them as one integrated system with one internal audit and one management review.
  • If enterprise buyers ask about AI only in a security questionnaire, a lighter AI governance engagement may answer them before a certification programme is needed.
ISO/IEC 27001, ISO/IEC 42001 and the EU AI Act side by side
AspectISO/IEC 27001ISO/IEC 42001EU AI Act
What it isA management system standardA voluntary management system standard; current edition ISO/IEC 42001:2023, published December 2023Law: Regulation (EU) 2024/1689
What it manages or governsInformation security riskAI-specific governance: the AI system inventory, impact assessment, data for AI, the AI lifecycle, transparency and third-party AI providersObligations that depend on your role, such as provider or deployer, and on each AI system's risk classification
StructureContext, leadership, risk process, internal audit, management review and corrective actionThe same management system structure, so the two can share one systemNot a management system; an ISO/IEC 42001 system keeps much of the evidence its duties depend on current
What a certificate showsConformity with the standard for the certified scope, issued by an accredited certification bodyConformity with the standard for the certified scope, issued by an accredited certification bodyNo certification body can grant compliance with the Act, and a 42001 certificate does not establish it

Which framework a startup should take on first is covered on the startup hub and in the help article on choosing a framework or report.

EU AI Act

How it relates to the EU AI Act

The EU AI Act, Regulation (EU) 2024/1689, is law. ISO/IEC 42001 is a voluntary management system standard. A 42001 certificate does not establish compliance with the Act, and no certification body can grant that.

What the management system does is make the evidence the Act's duties depend on exist and stay current: the inventory, the role for each system, risk management, data governance, documentation, human oversight and monitoring. The Act's obligations depend on your role and on each system's risk classification, which the EU AI Act page sets out, together with the dates that apply.

Where SecureRoot fits

What SecureRoot does in an ISO 42001 programme

SecureRoot's ISO 42001 work runs through scoping and the AI system inventory, AI risk and impact assessment, AI control design and the Statement of Applicability, implementation and monitoring, internal audit and management review, and support through the certification audit. The deliverables are an AI system inventory, AI impact assessment reports, an AI management system policy set, an AI control mapping and certification audit support.

Our published pricing row for ISO 42001: fee "Scoped, then quoted in writing"; third-party cost "The accredited certification body's two-stage audit, paid by you"; timeline "4 to 6 months to a first certificate". The pricing page shows the full row.

Scope boundaries

What this does not cover

  • SecureRoot is not an accredited certification body and does not issue ISO/IEC 42001 certificates. The certificate comes from the certification body you appoint, after its own audit.
  • No consultant can promise a certificate, and an ISO/IEC 42001 certificate is not compliance with the EU AI Act or any other law.
  • SecureRoot does not give legal advice. Classifying a system under the EU AI Act and interpreting its duties may need your legal counsel.
  • An AI management system does not test models for security flaws, bias or accuracy by itself. Your engineers run the evaluations and monitoring the controls require.
  • Your team's time building the inventory, running assessments and producing evidence is not inside any fee, and the certification body's audit fee is contracted and paid by you directly.
  • No model weights, training data, customer datasets or credentials are needed to scope the work.
  • This guide describes ISO/IEC 42001 in our own words. It does not reproduce the standard, which you should buy from ISO or your national standards body.

Questions

Common questions

  • Is ISO 42001 only for companies that build AI models?

    No. ISO describes it as applicable to organisations that develop, provide or use AI-based products or services. A startup building on a third-party model API is in scope, and so is governance of the AI tools its staff use.

  • Do we need ISO 27001 before ISO 42001?

    No, but they share a management system structure, so holding ISO/IEC 27001 makes ISO/IEC 42001 faster to add. If customers ask for both, build one integrated system.

  • Does an ISO 42001 certificate make us compliant with the EU AI Act?

    No. The Act is law and certification to a voluntary standard does not establish compliance with it. The management system produces much of the evidence the Act's duties need.

  • What is the difference between ISO 42001 and ISO 42005?

    ISO/IEC 42001 sets requirements for the AI management system and can be certified. ISO/IEC 42005 gives guidance on carrying out AI system impact assessments.

Primary Sources

Where the Figures on This Page Come From

Figures and requirements above are taken from the sources below, linked to the primary text wherever one exists rather than to a summary of it.

Related

Services and help that go with this

Governing AI in your product?

Tell SecureRoot which AI systems you build or rely on and who is asking about AI governance. We will come back with a scope and a written quote, with the certification body's fee shown separately.

Request an ISO implementation proposalSee ISO pricing