ISO 42001 vs NIST AI RMF vs SOC 2 (2026): What Buyers Ask For

The questionnaire came back with a new tab. Your biggest customer's procurement team added an AI section this renewal cycle. They want to know which models you use, whether customer data trains anything, who reviews model output, how you log inferences, and whether you hold "ISO 42001 or equivalent." Your SOC 2 Type 2 report answers none of that, and the person who wrote the questions probably knows it.

Three names keep coming up: ISO/IEC 42001, the NIST AI Risk Management Framework, and SOC 2 with AI controls added. They are not three versions of the same thing. One is a certifiable management system standard. One is a voluntary framework with no certificate at the end. One is an attestation report that was never written with AI in mind but is where most of your customers will look first. Picking the wrong one costs a year and a budget line. Picking the right one usually means picking two.

	ISO/IEC 42001	NIST AI RMF	SOC 2 (with AI controls)
What it is	International standard for an AI management system (AIMS), published December 2023	Voluntary US framework for managing AI risk, released January 2023	Attestation examination of controls at a service organization against the Trust Services Criteria
Scope	The organization's management system for developing, providing or using AI	Your AI risk practice: Govern, Map, Measure, Manage	A defined service system and its controls; AI enters through the system description
Who issues it	An accredited certification body (ANAB, UKAS, RvA and others)	Nobody. There is no certifier	A licensed CPA firm
What the customer receives	A certificate with a scope statement	Nothing formal; your own policies, inventory and risk register	A Type 1 or Type 2 report with the auditor's opinion
Evidence it demands	Clauses 4 to 10 plus a Statement of Applicability against 38 Annex A controls	Documented outcomes against the Playbook's suggested actions	Control evidence over the audit period; AI evidence mapped to existing criteria
Renewal	Three-year certificate, surveillance audits in years two and three	None	Annual report; Type 2 period usually 3 to 12 months
Best fit	Companies shipping AI features to buyers who ask for it by name	Any company that needs a defensible AI risk practice now	Any SaaS vendor selling to enterprise; the baseline most questionnaires are built on

ISO/IEC 42001: the only one that ends in a certificate

ISO/IEC 42001:2023 was published on 31 December 2023 and, in BSI's catalog wording, "specifies requirements and provides guidance for establishing, implementing, maintaining and continually improving an AI management system within the context of an organization." It uses the same clause 4 to 10 skeleton as ISO 27001. As SGS puts it: "ISO 42001, ISO 27001, and ISO 9001 all follow the same harmonized structure known as Annex SL." If you already run an ISMS, the management review, internal audit and corrective action machinery carries over.

Annex A holds 38 reference controls across nine areas. Schellman's overview describes Annex B as implementation guidance for those controls and Annex C as a catalog of objectives and risk sources. You select controls, justify exclusions in a Statement of Applicability, and the auditor tests against both the clauses and your SoA.

Accreditation caught up in stages. Schellman announced on 24 September 2024 that it was "the first ISO 42001 certification body accredited by ANAB." BSI announced on 17 November 2025 that it was "the first Certified Body in the world accredited by the United Kingdom Accreditation Service (UKAS)." Between them, ISO published ISO/IEC 42006:2025, the competence rules for the bodies doing the auditing.

The large AI vendors moved early, and their certificates are now what you get compared against. AWS announced on 25 November 2024 that it was "the first major cloud service provider to announce ISO/IEC 42001 accredited certification for AI services," covering Amazon Bedrock, Amazon Q Business, Amazon Textract and Amazon Transcribe. Google Cloud followed on 18 December 2024, Anthropic on 13 January 2025, and Microsoft 365 Copilot on 25 March 2025. OpenAI's security page states that it "maintains an ISO/IEC 42001:2023 AI Management System."

Two cautions. Schellman's FAQ says that after stage two, "organizations would be certified for a period of three years," with surveillance reviews in years two and three, and puts the stage one and stage two audit fee "from the $20,000's up to the $40,000's" with surveillance at "$13,000 to $20,000+ annually." That is the audit fee only; implementation is scoped per engagement. Second, ISO 42001 is not a shortcut to EU AI Act compliance. The Cloud Security Alliance's April 2026 research note is blunt: "ISO/IEC 42001 alone does not satisfy the EU AI Act, despite frequent industry shorthand to the contrary."

NIST AI RMF: the practice, without the paper

NIST released AI RMF 1.0 on 26 January 2023. It is, in NIST's words, "intended for voluntary use." Nobody certifies against it. What you can hand them is your own work across the four functions: Govern assigns ownership and policy, Map inventories systems and context, Measure tests and tracks, Manage treats what Measure finds. NIST's Core describes Govern as cultivating "a culture of risk management within organizations designing, developing, deploying, evaluating, or acquiring AI systems." The word acquiring is why the framework fits companies that only use AI tools.

The companion Playbook, first published in full on 30 March 2023, lists suggested actions per subcategory and tells organizations to borrow as many or as few as apply. For generative AI, NIST released the Generative AI Profile, NIST AI 600-1, on 26 July 2024. It names twelve generative AI risks. Our NIST AI RMF implementation guide walks through standing up each function at mid-market scale.

Regulatory references to the framework move. Colorado's 2024 AI Act, SB 24-205, gave deployers an affirmative defense for a risk program complying with frameworks "such as" the NIST AI RMF or ISO 42001, per Gibson Dunn's summary. Then the legislature replaced the statute: SB 26-189, signed 14 May 2026 and effective 1 January 2027, removed "requirements to perform annual impact assessments and maintain a risk management program." Build the practice because customers and your own engineers need it, not because a statute names it this quarter.

SOC 2: where the AI questions actually land

SOC 2 is an attestation, not a certification. A licensed CPA firm examines the controls around a defined system against the Trust Services Criteria (Security, Availability, Processing Integrity, Confidentiality, Privacy) and issues an opinion. A Type 1 covers control design at a point in time; a Type 2 covers design and operating effectiveness over a period, usually 3 to 12 months, as A-LIGN's guide lays out. Enterprise buyers already know how to read one, which is why AI assurance lands here first, even though the criteria never mention AI.

The Cloud Security Alliance's August 2026 note records that the AICPA "has not published AI-specific Trust Services Criteria, AI-specific points of focus, or an authoritative AI addendum," that the criteria were "last substantively revised in 2017, with a points-of-focus refresh in 2022," and that auditors are building evidence requests and "mapping them back to the unchanged 2017 Trust Services Criteria." Two auditors can scope the same AI feature differently, so agree the AI evidence list with yours before the period starts.

AI enters a SOC 2 through three doors. The system description: if your product has an AI feature, it belongs in the description, and the controls around it get tested. The criteria mapping: Baker Tilly's December 2025 guidance maps AI risk into the existing COSO components. The subservice organizations: "When AI solutions depend on third-party cloud providers, external datasets, or APIs, SOC 2 includes criteria for assessing and managing risks arising from relevant third-party service providers." Your model vendor is a subservice organization, and its ISO 42001 certificate is part of your evidence. Our SOC 2 compliance services cover that scoping without inflating the audit.

What customers are actually asking for

Look at the instruments procurement teams copy from. The Cloud Security Alliance published its AI Controls Matrix on 10 July 2025 with "18 security domains containing 243 control objectives," plus a CAIQ for AI described as "an evaluation tool for third-party vendors." Shared Assessments' SIG, the questionnaire behind many bank and insurer vendor reviews, said of its 19 September 2025 release that "the SIG now references ISO 42001." HECVAT 4, released by EDUCAUSE in February 2025 for higher education buyers, "includes a set of AI-specific questions developed by community volunteers."

Survey data points the same way. Vanta's State of Trust report, from 2,500 IT and business leaders, found that "nearly two-thirds (65%) of organizations say that customers, investors, and suppliers are increasingly requiring proof of compliance," while only 36% had or were putting an AI policy in place. The buyer side is asking faster than the vendor side is documenting.

Notice what the questions ask for: an AI policy, an inventory, a human review step, data-use terms with your model vendor, logging, incident handling. They rarely demand a certificate on the first pass. "ISO 42001 or equivalent" signals that the buyer wants a management system; a documented NIST-aligned practice plus a SOC 2 that names the AI feature in its system description is the equivalent most of them accept. Our AI vendor risk questions for CISOs are the mirror image.

A decision rule by company type

SaaS selling to enterprise

Keep SOC 2 Type 2 as the spine. Add the AI feature to the system description, agree the AI evidence overlay with your auditor, and run the NIST functions internally so the questionnaire answers are true. Put ISO 42001 on the roadmap when a named customer or a deal of real size requires the certificate.

Regulated industry: fintech, healthcare, hospitality with payment data

SOC 2 with AI controls remains the first ask. If you already hold ISO 27001, the Annex SL structure makes ISO 42001 the cheapest certificate you will ever add: the management system clauses are already running, and what you add is the AI impact assessment, the Annex A controls, and the SoA.

Shipping AI features versus using AI tools

If AI is in your product, the customer's risk is your model, your prompts, your data handling and your vendor chain: SOC 2 scoping plus a documented practice now, ISO 42001 when a customer names it. If AI is only in your workflow (Copilot, coding assistants, support bots), the customer's risk is your employees' data handling. An AI acceptable use policy, an approved-tool inventory and the Govern and Map functions cover nearly every question. Our AI governance checklist for mid-market companies is sized for this group. Treat agents separately: the AI Agent Security Report 2026 covers the identity and tool-access failures none of the three frameworks was written to catch.

The honest answer for most mid-market companies

Most mid-market companies need SOC 2 with the AI feature mapped in, and a documented AI risk practice built on the NIST functions, before they buy an ISO 42001 certification. It is also a management system on a three-year audit cycle, and a management system is only worth certifying once it exists. Build the practice, name the AI system in your SOC 2, and let a customer's written requirement set the timing of the certificate.

If you want a second set of eyes on which of these your customers are really asking for, BlueRadius runs an AI and Agent Security Assessment that inventories your AI systems and agents, maps them against all three, and returns a plan you can hand to procurement. If the AI section is one tab in a larger questionnaire, the fixed-fee posture assessment covers the whole thing.

Frequently Asked Questions

Is ISO 42001 a replacement for SOC 2?

No. ISO 42001 certifies an AI management system at the organization level; SOC 2 is an attestation report on controls over a defined service system. US enterprise buyers still ask for SOC 2 first.

Can we get certified against the NIST AI RMF?

No. NIST describes the framework and its Playbook as "intended for voluntary use," and no body certifies against it. What you can produce is a documented practice across Govern, Map, Measure and Manage.

Does ISO 42001 certification make us EU AI Act compliant?

No. The Cloud Security Alliance's April 2026 note states that "ISO/IEC 42001 alone does not satisfy the EU AI Act." The harmonized standard that would carry a presumption of conformity, prEN 18286, was still in process with publication targeted for late 2026.

How does our model provider's certification help us?

In a SOC 2, your model vendor is a subservice organization, and its ISO 42001 certificate (AWS, Google Cloud, Microsoft, Anthropic and OpenAI all publish one) is evidence for your vendor management controls. It does not cover your prompts, your data handling or your human review step.

What does ISO 42001 certification cost?

Schellman, the first ANAB-accredited certification body, publishes an audit fee range "from the $20,000's up to the $40,000's" for stage one and stage two, with surveillance at "$13,000 to $20,000+ annually." Implementation effort and timeline are scoped per engagement.