ISO 42001 vs ISO 27001: What's Actually New in an AI Management System

A practical breakdown of what ISO 42001 adds beyond ISO 27001 and which parts of an existing ISMS already meet AI Management System requirements.

If you already hold ISO 27001, I get why the first instinct with ISO 42001 is to reuse everything: the internal audit checklist, the management review agenda, and even the risk methodology. Some of that instinct is right. The mistake is stopping there.

ISO 42001 borrows the same high-level structure every recent ISO management-system standard uses: Clauses 4 through 10: context of the organization, leadership, planning, support, operation, performance evaluation, and improvement. If you're already running an ISMS, you already have a working internal audit process, a management review cadence, document control practices, and a risk methodology with the mechanics already worked out. None of that needs rebuilding from scratch.

Where the two standards actually part ways is Annex A. ISO 27001's Annex A is about information security controls: access control, cryptography, physical security, and incident response. ISO 42001's Annex A is a different animal entirely, built around risks specific to AI: system impact assessments, data quality and provenance for the data that trains and feeds the system, transparency and human oversight that's real rather than symbolic, responsible design and development processes, and a distinct layer of due diligence for third-party AI suppliers.

None of that shows up in an ISO 27001 audit, because it was never designed to. So if your plan for building an AIMS is copying your ISMS checklist and swapping "information security" for "AI" wherever it appears, you'll sail through the parts that genuinely overlap and miss everything that makes ISO 42001 its own standard.

The way I'd actually approach this: treat ISO 42001 as an extension of the ISMS you already have, not a duplicate of it. Share the governance scaffolding wherever your scope allows: leadership commitment, audit cadence, document control. But the content inside that scaffolding, especially Annex A, needs to be built specifically for AI risk. It can't be adapted line by line from a security checklist and still hold up.

If you're scoping an ISO 42001 build on top of an existing ISO 27001 program, the fastest way to avoid wasted effort is deciding early — before you write a single control — which parts are genuinely shared and which ones need their own AI-specific version.