ISO 42001 vs ISO 27701: Managing AI and Personal Data Under One System

ISO 42001 manages how you govern AI. ISO 27701 manages how you protect personal data. If your AI systems use personal data, and most do, you need the controls of both. The good news: since ISO 27701 was revised in October 2025 as a standalone standard, the two now share the same management-system structure, so you can run them as one integrated system with one risk process, one audit programme and one management review.

This guide explains what each standard covers, exactly where they overlap, what each one misses, and how to build a single system that can be certified against both.

What each standard is for

The overlap is obvious once you look at what AI runs on. Training data, prompts, outputs and logs often contain personal data. ISO 42001 asks whether the AI is responsible. ISO 27701 asks whether the personal data inside it is handled lawfully and fairly. Neither answers the other's question.

The 2025 change that makes integration easier

Until October 2025, ISO 27701:2019 was an extension of ISO 27001. You could not certify a privacy system without first holding, or certifying alongside, an information security management system. The 2025 edition removed that dependency.

ISO 42001 vs ISO 27701 at a glance

Where they overlap

The management system: almost identical

Both standards follow the ISO harmonised structure. That means the same clause headings, and in practice much of the same evidence.

The controls: where AI and privacy meet

At control level, four areas carry most of the overlap:

Which ISO 27701 controls apply depends on whether you act as a controller, a processor or both. The shared controls apply in every case.

Where each standard leaves a gap

The gaps matter in practice. In December 2024, Italy's data protection authority fined OpenAI €15 million over ChatGPT. Its findings included processing personal data for training without first identifying an adequate legal basis, and weak transparency. Those are privacy-management failures that an AI governance system on its own would not prevent.

The reverse is just as common. IBM's 2025 Cost of a Data Breach report found that 63% of the 600 organisations it studied had no AI governance policies. A privacy programme without AI governance has no answer to bias, model drift or misuse.

Which one do you need?

The timeline to plan around

• Aug 2019

  ISO/IEC 27701:2019 published as an extension to ISO 27001.

• Dec 2023

  ISO/IEC 42001 published, the first certifiable AI management system standard.

• 2025

  ISO/IEC 42006 published, setting requirements for bodies that certify ISO 42001.

• 14 Oct 2025

  ISO/IEC 27701:2025 published as a standalone standard; ISO/IEC 27706 replaces the old certification-body requirements.

• Oct 2028

  End of the transition: ISO 27701:2019 certificates expire.

How to build one integrated system: 6 steps

1. Set one scope covering AI and PII

   Define which AI systems and which personal data processing are in scope. Record your AI roles and your controller and processor roles together in Clause 4.

2. Combine policies, keep both commitments

   One top-level policy can cover AI and privacy, as long as it includes the specific commitments each standard requires. Name one accountable owner for each.

3. Run one risk process with two lenses

   Use one methodology and one register, with AI risks such as bias, misuse and drift alongside privacy risks such as unlawful processing, over-retention and unauthorised disclosure.

4. Merge impact assessments

   Build one template covering the AI system impact assessment and the privacy impact assessment. Add a section that can serve as a GDPR DPIA where required.

5. Build one combined Statement of Applicability

   List ISO 42001's 38 controls and the ISO 27701 controls that fit your roles in a single document, with justifications for inclusions and exclusions.

6. Audit and review once

   Run one internal audit programme and one management review covering both standards. Ask your certification body about a combined audit to cut days and cost.

Common mistakes to avoid

• Building two separate management systems with two risk registers
• Assuming you still need ISO 27001 before ISO 27701
• Forgetting that ISO 27701:2019 certificates expire in October 2028
• Treating AI training data as purely a quality issue, ignoring its lawful basis
• Running privacy impact assessments that never mention bias or AI outputs
• Claiming certification means GDPR or AI Act compliance

A prompt to draft your integrated Statement of Applicability

Paste in your context to get a first-draft combined SoA. Have your lead implementer or auditor review it before use.

You are a lead implementer for ISO/IEC 42001:2023 and
ISO/IEC 27701:2025, building one integrated management system.

Our organisation:
[what we do, AI systems in scope, personal data we process,
whether we act as PII controller, processor or both, countries
we operate in, existing certifications]

Produce:
1. A combined scope statement covering the AI systems and the PII
   processing in scope, with our AI roles and privacy roles.
2. A Statement of Applicability table with columns:
   - Standard (42001 or 27701)
   - Control reference and title
   - Applicable? (yes/no)
   - Justification
   - Shared evidence that also satisfies the other standard
3. A list of controls where one piece of evidence covers both
   standards, grouped by theme (data, impact assessment,
   transparency, third parties).
4. The ten controls most likely to need new work, ranked by effort.

Use only the facts I provide. Mark anything uncertain as UNKNOWN.
Do not invent control text; refer to controls by reference and
short title only.

ISO 42001 asks whether your AI is responsible. ISO 27701 asks whether the data inside it was handled lawfully. Run them as one system and you answer both questions with one set of evidence. — Kumail Mehdi

Frequently asked questions

Sources

• ISO — ISO/IEC 42001:2023 AI management systems
• A-LIGN — ISO 27701 updates: what you need to know
• ISMS.online — ISO 27701:2025 vs 2019 comparison
• Lewis Silkin — Garante fines OpenAI €15 million
• IBM — 2025 Cost of a Data Breach: navigating AI
• The Register — DLA Piper GDPR fines survey 2026
• Stepto — ISO 42001 certificate count
• CEPA — What counts as safe AI (ISO/IEC 42006)