ISO 42001 vs SOC 2: Which One Do Enterprise AI Buyers Actually Ask For?

Short answer: enterprise buyers still ask for SOC 2 first, because it is already written into their contracts and vendor reviews. But if AI is part of what you sell, ISO 42001 is the question that now follows straight after it — and in regulated and EU-facing deals, it is increasingly the one that decides who makes the shortlist.

If you sell software with AI features into large organisations, you have probably seen both acronyms land in the same security questionnaire. This guide explains what each one actually proves, what the data says about how often buyers ask for them, and how to decide which to pursue first — without spending a year and a six-figure budget on the wrong badge.

What SOC 2 and ISO 42001 actually prove

The two are often lumped together as "compliance badges", but they answer completely different questions a buyer has about you.

One important wording difference buyers notice: there is no such thing as a "SOC 2 certificate". SOC 2 is an attestation report. ISO 42001, by contrast, is a genuine certification issued by an accredited certification body. Saying "we're SOC 2 certified" in a sales deck is a small but telling red flag to experienced procurement teams.

ISO 42001 vs SOC 2 at a glance

So which one do enterprise buyers actually ask for?

Here is the honest answer, based on the numbers rather than vendor marketing: SOC 2 is the requirement. ISO 42001 is the rising question.

SOC 2 is already in the contract

Third-party audit reports are now the standard way companies prove security to buyers. Secureframe's 2026 benchmark found that 73% of companies share a third-party audit report such as SOC 2, and 70% also complete security questionnaires or RFPs. In ISC2's supply-chain research, 77% of respondents listed compliance with standards like SOC 2, ISO 27001 or NIST among their vendor security requirements, and 71% required vendors to provide audits, attestations or assessments.

That last figure matters. Even with a clean SOC 2 report, most buyers keep asking questions — and for AI vendors, the follow-up questions are increasingly about the AI itself.

ISO 42001 is the question that comes next

ISO 42001 adoption is still small but moving fast. Public tallies built from certification-body and company announcements put the global total at roughly 350 certified organisations in spring 2026. For scale, the 2024 ISO Survey recorded 96,709 valid ISO/IEC 27001 certificates. One audit firm's analysis found that across three mainstream CPA firms serving around 11,000 clients, only about 60 were pursuing ISO 42001 as of June 2026 — roughly half of one percent.

So why does it keep appearing in RFPs? Because the companies at the top of the AI supply chain hold it. AWS became the first major cloud provider with accredited ISO 42001 certification in November 2024. Anthropic announced its certification, issued by ANAB-accredited Schellman, in January 2025. Google, Microsoft, OpenAI and IBM have since certified parts of their AI operations. Once the biggest suppliers can answer a question, procurement teams start asking it of everyone.

Buyer behaviour is shifting too. An industry analysis of responsible-AI governance data found ISO/IEC 42001 cited as an influence by 36% of organisations in its first year of tracking, while the share of organisations with no responsible-AI policy at all fell from 24% to 11%. Governance has moved out of the ethics committee and into the vendor questionnaire.

Why SOC 2 alone no longer covers AI

The gap is structural. The Trust Services Criteria behind every SOC 2 report were last substantively revised in 2017, with a points-of-focus refresh in 2022 — before generative AI changed enterprise software. The AICPA has confirmed that AI does not get a separate SOC examination or new criteria; it is assessed under whichever existing categories are in scope.

The Cloud Security Alliance noted in August 2026 that, without AI-specific criteria, auditors are improvising their own evidence requests around model lineage, prompt and inference logging, drift monitoring and AI sub-processor risk. That means two SOC 2 reports from two AI vendors can test very different things.

Scope makes it worse. In one CBIZ benchmark sample, 12 of 73 SOC 2 reports covered Security alone — no Processing Integrity, which is the category most relevant to whether an AI system's outputs are reliable.

What buyers really ask AI vendors in 2026

The certificate opens the door. The questions behind it are what slow deals down. AI-specific questionnaire sections now commonly ask vendors to:

• Provide an inventory of AI systems and the date it was last reviewed
• Show an AI risk and impact assessment for a production system
• Explain whether customer data is used to train or fine-tune models, and when it is deleted
• Name the person accountable for AI decisions and the AI incident escalation path
• Describe controls over third-party model providers and AI sub-processors
• Classify the system under the EU AI Act risk tiers

Every one of these maps directly to an ISO 42001 clause or Annex A control — which is exactly why the standard is useful even before you certify. The Cloud Security Alliance also released its AI-CAIQ questionnaire in October 2025, giving buyers a ready-made AI controls checklist to send to vendors.

Unanswered gaps are expensive. One 2026 vendor guide estimates each unaddressed questionnaire gap adds two to four weeks of remediation and legal review, and that six gaps can push a deal back three to six months.

Where the EU AI Act fits

The EU AI Act is pulling ISO 42001 into European procurement, but it is easy to overstate the link. Under the Digital Omnibus agreement reached in May 2026, obligations for stand-alone high-risk AI systems (Annex III) move to 2 December 2027, and those for AI embedded in regulated products (Annex I) move to 2 August 2028.

Which should you pursue first?

For most AI vendors, the right answer is sequence, not choice: SOC 2 as the baseline, extended to cover your AI components, then ISO 42001 as the AI layer. The evidence overlaps heavily — access control, change management, supplier management and incident response all carry across — so the second one costs far less than the first.

What it costs and how long it takes

A practical 4-step path for AI vendors

1. Ask your buyers what they need

   Before buying any audit, ask your top five prospects' security teams what assurance they expect over your AI features: a framework, specific controls, or questionnaire answers. Their answer decides your order.

2. Get SOC 2 right — with AI in scope

   Name your models, inference endpoints, training data stores and AI sub-processors in the system description. Add Processing Integrity if customers rely on your AI outputs.

3. Build the ISO 42001 artefacts before the audit

   An AI system inventory, an AI policy signed by an accountable owner, risk and impact assessments, and a Statement of Applicability. These answer most AI questionnaire sections on their own.

4. Certify when the pipeline justifies it

   Once regulated or EU deals are in play, take the management system to a properly accredited certification body. Check who accredited the auditor before you sign.

A prompt to answer AI governance questionnaires faster

Use this with your own documentation pasted in. It turns scattered policies into consistent, honest answers — and shows you where the gaps are.

You are an AI governance lead preparing responses to an enterprise
customer's AI vendor security questionnaire.

Context about our company and product:
[paste your AI policy, system inventory, SOC 2 system description,
data handling and sub-processor list]

The questionnaire questions:
[paste the questions]

For each question:
1. Write a clear, factual answer of no more than 120 words using only
   the context provided. Do not invent controls we have not described.
2. Name the evidence we could attach (policy, log, assessment, report).
3. Map the answer to the closest ISO/IEC 42001 clause or Annex A control
   and, where relevant, the SOC 2 Trust Services Criterion.
4. If the context does not support an answer, write "GAP" and describe
   what we would need to build to answer it honestly.

Finish with a table of all GAP items, ranked by how likely they are to
block an enterprise deal.

SOC 2 tells a buyer your doors are locked. ISO 42001 tells them someone is responsible for what the AI does once it is inside. — Kumail Mehdi

Frequently asked questions

Sources

• Secureframe — SOC 2 vs security questionnaires (2026 benchmark)
• Bright Defense — SOC 2 statistics for 2026 (ISC2, Whistic and CBIZ figures)
• Stepto — ISO 42001 holder count and 2024 ISO Survey comparison
• Throughline — ISO 42001 for AI companies
• CoreLine — ISO 42001, NIST AI RMF and the EU AI Act: what buyers now ask
• Cloud Security Alliance — The SOC 2 AI gap (August 2026)
• Schneider Downs — AICPA TQA Section 9561 on AI in SOC examinations
• Anthropic — ISO 42001 certification announcement
• CEPA — What counts as safe AI
• Gibson Dunn — EU AI Act Omnibus agreement
• Expert Insights — SOC 2 compliance guide 2026 (timeline and cost)
• Promise Legal — AI procurement questionnaires and AI-CAIQ