ISO 42001 AI Management System Certification

Any organisation that develops, provides or relies on AI systems and needs to show it governs them responsibly. In practice the demand comes from three directions. Enterprise customers now ask how AI features in your product are trained, monitored and overseen, often in the same questionnaire as SOC 2 and ISO 27001. Regulated buyers need assurance before they can use your system in their own risk-managed process. And companies facing the EU AI Act want a management system that makes those duties repeatable rather than ad hoc. It is equally relevant to organisations that only deploy third-party AI, since governance of what you buy, how staff use it and what data it sees is a large part of the standard. For companies that already hold ISO 27001 or SOC 2, the incremental work is usually smaller than expected, because the management system exists and only AI-specific governance is added. That makes certification a faster answer to buyer questions than building a bespoke AI policy set nobody recognises.

The standard is a management system; the Act is law. ISO 42001 gives you the governance machinery the Act's duties need: an inventory of AI systems, risk and impact assessment, data governance, documentation, human oversight, monitoring after deployment and supplier controls. Certification is not compliance with the Act, and no certification body can grant that. The Act's obligations depend on your role as provider or deployer and on the risk classification of each system, and conformity assessment for high-risk systems follows its own route. What ISO 42001 does is make the evidence exist and stay current, so the Act's technical documentation, oversight and monitoring requirements are produced by a working system rather than assembled in a rush. We map the standard's controls to the specific duties your systems carry. We also keep the mapping current, because the Act's timetable has already changed once and harmonised standards are still arriving, and a governance system that tracks those changes is worth more than a one-off gap report.

Yes. You define the scope of the management system, so it can cover one product, one business unit or the whole organisation, as long as the boundary is coherent and defensible. Starting narrow is common and sensible: it gets the governance working around the AI that matters most commercially, proves the model to customers and auditors, and creates patterns you can extend. The scope statement has to be honest about what is inside and outside, because your certificate names it and a customer will read it. Where AI systems share data pipelines, models or teams, drawing the line too tightly can create more work than it saves. We help you choose a boundary that is meaningful to buyers and practical to run, then widen it in later cycles. We also document what sits outside the scope and why, so a customer reading the certificate understands its limits rather than assuming it covers every model you run. That honesty avoids a difficult conversation later in procurement.

It looks at what an AI system does to people and organisations affected by it, not just at risks to your business. A typical assessment records the system's purpose and context, the individuals or groups affected, the data used to train and run it, foreseeable harms such as unfair outcomes, inaccuracy, privacy intrusion or over-reliance, the controls in place, and the residual risk someone accountable accepts. It also records human oversight: who can review, override or switch the system off. This is the document that most often does not exist before we arrive, and it is the one auditors, enterprise buyers and EU AI Act obligations all lean on. We run assessments with the product and data teams, so conclusions reflect how the system actually behaves. We also set a review trigger, so an assessment is repeated when the model, its data or its purpose changes materially rather than only at audit time. A stale assessment is treated by auditors as no assessment at all.

For most organisations, four to six months to a first certificate, following our phases. Scoping and the AI inventory take one to two weeks, risk and impact assessment two to three, control design three to four, implementation and monitoring four to eight, and the internal audit and management review one to two, before the certification body's audit. Companies that already hold ISO 27001 move faster, because the management system structure, risk process and audit routine exist and only AI-specific controls are added. What stretches the timeline is usually the inventory: teams routinely find AI in features, vendor tools and internal workflows nobody had catalogued. We confirm a timeline after the inventory and risk assessment, when the number of systems in scope is known. Booking the certification body early helps, since accredited bodies offering ISO 42001 are still fewer than for ISO 27001 and audit slots can be several weeks out. We plan that window with you during scoping.

An accredited certification body issues it after auditing your AI management system, and it must be independent of whoever built that system. SecureRoot is not a certification body; we build the system and support you through the audit. We do not publish a price, because scope drives the effort: how many AI systems are in scope, whether you are a provider, a deployer or both, how much data governance already exists, whether an ISO 27001 management system is available to build on, and how many impact assessments are needed. The certification body's fee and your team's time sit outside our fee. Pursuing ISO 42001 alongside ISO 27001 lowers the total, because the management system is shared. We scope first, then quote a fixed price in writing. Where your buyers are asking about AI in a security questionnaire rather than demanding a certificate, we will say so, and scope a lighter governance engagement instead of a certification programme you do not yet need.