AI products · SaaS · Platforms · AI agents

When customers start asking

ISO 42001 for
AI and Software Companies

Enterprise customers are adding ISO 42001 to vendor security reviews. This page covers what they actually want, how much of it your ISO 27001 or SOC 2 program already gives you, the AI-specific controls you will need to build, and how to decide between documented alignment and a certificate.

"Keep it Simple. Keep it Real."

Schedule Free Consultation 858-240-4353
27001 or SOC 2
Build on what you have
Annex A
The AI-specific part
$9,750
Fixed-fee gap assessment
Jared Clark, ISO 42001 consultant for AI and software companies

What Customers Want

The question arrives in a security review

For most AI and software companies, ISO 42001 shows up as a line in an enterprise customer's security questionnaire or vendor risk review: are you certified, and if not, how do you govern the AI in your product? The first few hundred organizations to certify were mostly AI platforms and software companies, and procurement teams noticed. Insurers and investors have started asking the same question in their own words.

What the customer actually needs varies more than the question suggests. Some want a certificate from an accredited certification body. More often they want documented alignment: an AI policy, an inventory of the AI in your product and your operations, impact assessments for the features that affect their users, and evidence that the controls run. The honest first step is to ask the customer which one they need, because the difference is months of work.

In our view the useful way to see ISO 42001 is as a management system wrapped around the engineering you already do. Your team already reviews code, tests releases and watches production. The standard asks you to make those habits deliberate for AI, write down who owns them, and be able to show a stranger that they happen.

What You Already Have

ISO 27001 carries over more than SOC 2

Both help. They help in different places, and planning on the wrong one is how projects run long.

ISO 42001 needsFrom ISO 27001From SOC 2
Management system: policy, objectives, management review, internal audit, improvementLargely in place; extend to AIMostly new
Risk assessment and treatmentMethod in place; add AI risksPartial; add a documented method
Statement of ApplicabilityFamiliar; write a new one for Annex A of ISO 42001New
Access, change management, vendor review, loggingIn place; extend to models and promptsIn place; extend to models and prompts
AI system inventory and AI impact assessmentsNewNew
AI lifecycle, data provenance, information for usersNewNew

Companies with a working ISO 27001 system are often at the short end of the four-to-twelve-month range to certification. The clause-level detail is in ISO 42001 clause by clause and from ISO 27001 to ISO 42001.

What Is New

The controls for AI you ship

This is where a security program is usually thinnest, and where customers' reviews have started to probe.

Model and prompt change control

A model upgrade, a new system prompt or a changed retrieval source can change behavior as much as a code release. Treat them as releases: reviewed, tested against known cases, approved and recorded.

Data provenance

Where training, fine-tuning and evaluation data came from, what rights you have to it, and whether customer data is used. Customers ask this first, and the answer has to be a record rather than a recollection.

Model providers as suppliers

If you build on a foundation model API, that provider is part of your AI system. What the contract says about data retention, training on your inputs and change notice belongs in your supplier controls.

Impact assessments for features

For each AI feature that affects your customers' users: who could be harmed, how, and what you do about it. Short, specific and revisited when the feature changes.

Monitoring in production

How you would notice a model behaving differently this month than last: evaluation sets that run on a schedule, alerts on output quality, and an incident path that treats AI failures as incidents.

Information for users

What you tell customers about how the AI works, its limits and their responsibilities. Annex A expects it, and enterprise legal teams increasingly write it into contracts.

AI Agents

If your product uses agents

ISO 42001 governs the organization and its AI systems, and its controls do apply to agents, but it was written before tool-using agents were common. Customers buying agent products now ask about things the standard does not spell out: which tools an agent may call, what it can do without a human approving it, how delegation between agents is limited, and how you would reconstruct what an agent did after the fact.

For agent products we add those controls inside the same management system and map them to AIUC-1, an assurance standard written specifically for AI agents that publishes a crosswalk to ISO 42001. The two work together: ISO 42001 at the level of the organization, AIUC-1 at the level of the agent.

How We Work

Alongside your security team

We do not replace your security program or your assessor. We bring the management-system discipline ISO 42001 requires, which is the part security programs most often lack, and we build it into the policies, tickets and release process your engineers already use so it does not become a second bureaucracy.

Our background is management systems and regulated industries: quality, validation and regulatory work where every change to a system has to be justified and recorded. That habit is what an ISO 42001 auditor is looking for, and it is also why regulated enterprise customers trust vendors who have it. We build and run AI systems of our own, so the lifecycle controls we recommend are ones we have had to live with.

A certification body cannot consult on what it certifies, so the firm that helps you build the system and the firm that audits it are always different. We will help you choose an accredited body and prepare for both audit stages.

Where to start

  1. Ask the customer whether they need a certificate or documented alignment. It decides the size of everything that follows.
  2. Run a gap assessment. Ours is a fixed $9,750: every clause and Annex A control, what your ISO 27001 or SOC 2 program already covers, a prioritized plan and an estimate of time and cost. What it covers.
  3. Build what is missing inside the tools your team already uses, then run an internal audit and management review.
  4. Book the certification body early if you are certifying. Auditor availability has been the long pole in 2026.

Selling Into Europe

ISO 42001 is a foundation for the EU AI Act, not a pass

The European adoption of ISO 42001 was published in March 2026 but has not been cited in the Official Journal, so certification gives no presumption of conformity with the AI Act. The standard being written for the Act's quality management requirements is prEN 18286. If you are a provider of a high-risk system under the Act, ISO 42001 gives you much of the governance the Act expects, and you will still need the Act's own conformity route.

The detail is in ISO 42001 and the EU AI Act and ISO 42001 vs NIST AI RMF vs the EU AI Act. Costs and timelines are in what ISO 42001 costs and how long certification takes.

Frequently asked questions

A customer asked if we are ISO 42001 certified. Do we need the certificate?

Not always. Many enterprise security reviews accept documented alignment with ISO 42001: an AI policy, an inventory of AI systems, impact assessments and the controls you run, with evidence. A certificate becomes worth it when several large customers require it by name, when it is written into contracts, or when it helps you win against competitors who have one. Ask the customer which they need before you commit.

We have SOC 2. How much of ISO 42001 does that cover?

Less than people expect. SOC 2 gives you control evidence, access and change management, vendor review and an audit habit, all of which help. It does not give you the management-system clauses ISO 42001 requires, such as an AI policy with objectives, management review, internal audit and continual improvement, and it does not cover AI-specific risks. Plan to build more than you would from ISO 27001.

Does ISO 42001 cover AI agents?

Partly. ISO 42001 governs the organization and its AI systems, and its controls apply to agents, but it was not written with tool-using agents in mind. For agent products we add controls for tool authorization, delegation and runtime monitoring, and we map them to AIUC-1, an assurance standard for AI agents that publishes a crosswalk to ISO 42001.

Will ISO 42001 satisfy the EU AI Act for our product?

No, not on its own. The European adoption of ISO 42001 has not been cited in the Official Journal, so it gives no presumption of conformity with the AI Act. The standard being written for the AI Act's quality management requirements is prEN 18286. ISO 42001 is still a strong foundation for the governance the Act expects of providers.

Get a straight answer before the next questionnaire

Book a free 30-minute call. Tell us what your customers are asking for and what you already have in place, and we will tell you whether documented alignment will do or a certificate is worth it, and what either would take.

Schedule Free Consultation Call 858-240-4353

Or email us at [email protected]