Short answer
SOC 2 Type 2 holders reuse about 40–50% of their evidence base when adding ISO 42001. (LowerPlane practitioner estimate, based on running both frameworks for ~30 AI startups through 2026.) The reusable half covers management-system scaffolding (CC1 → clauses 4–10) and information-security controls (CC6, CC8, CC9). The AI-only half is Annex A controls A.5 through A.9 — impact assessment, model lifecycle, dataset data-sheets, transparency and use-case governance. No SOC 2 shortcut covers those.
Crosswalk: CC1–CC9 ↔ clauses 4–10 and Annex A
| SOC 2 criterion | ISO 42001 clause | ISO 42001 Annex A | Reuse notes |
|---|---|---|---|
| CC1 — Control Environment | Clauses 4, 5 (Context, Leadership) | A.2 (Policies), A.3 (Internal organization) | Governance structure, named owner, policy pack — direct reuse. |
| CC2 — Communication & Information | Clause 7.4 (Communication) | A.8 (Information for interested parties) | Transparency obligations are AI-specific; SOC 2 CC2 covers the general case. |
| CC3 — Risk Assessment | Clause 6.1 (Risk management) | A.4 (Resources for AI systems), A.5 (Impact assessment) | Risk register reuses; AI impact assessment is new. |
| CC4 — Monitoring Activities | Clause 9 (Performance evaluation) | A.6.2.6 (AI system operation and monitoring) | Monitoring cadence reuses; drift and bias monitors are AI-specific. |
| CC5 — Control Activities | Clause 8 (Operation) | A.6, A.7 | Change management reuses; model and data lifecycle are new. |
| CC6 — Logical & Physical Access | (Inherits ISO 27001) | (Inherits ISO 27001) | Full reuse — ISO 42001 references ISO 27001 for access control. |
| CC7 — System Operations | Clause 8.3 (Operational planning) | A.6.2.6 (operation & monitoring) | Incident response reuses; prompt injection and drift scenarios are AI-specific. |
| CC8 — Change Management | Clause 8.2 (Operational planning) | A.6.2 (AI system development) | Change workflow reuses; model-version pinning and prompt review are AI-specific. |
| CC9 — Risk Mitigation (Vendor) | Clause 8.3 | A.10 (Third-party relationships) | Vendor-review cadence reuses; LLM-provider DPA and no-training clause are AI-specific. |
This mapping is a practitioner view; confirm against your auditor's preferred mapping before using it for formal attestation.
Controls that are AI-only
The five Annex A control objectives below have no SOC 2 analogue in the current (2017, revised 2022) Trust Services Criteria. They are the work you cannot shortcut.
A.5 — AI System Impact Assessment
What it covers: Document intended use, affected individuals/groups, foreseeable harms, benefits, mitigations. Review on material change.
Evidence an auditor samples: AI impact assessment template per production model
A.6 — AI System Life Cycle
What it covers: Stages: planning, data, design, verification, deployment, operation, decommissioning. Controls at each boundary.
Evidence an auditor samples: Evaluation report per release, deployment approval record, decommissioning attestation
A.7 — Data for AI Systems
What it covers: Dataset provenance, quality metrics, acquisition licence, consent basis. Separate for training, validation, operation.
Evidence an auditor samples: Dataset data-sheet per corpus
A.8 — Information for Interested Parties
What it covers: Transparency to users and affected parties. User disclosure of AI usage, limitations, points of contact.
Evidence an auditor samples: Model card, user-facing disclosure copy, affected-parties register
A.9 — Use of AI Systems
What it covers: Documented intended use, use-case governance, human oversight, redress mechanism.
Evidence an auditor samples: Intended-use statement, oversight runbook, incident log
How the 2027 TSC AI points of focus narrow the gap
The AICPA's 2027 Trust Services Criteria revision adds 7 AI points of focus under existing CC-series criteria. Four of them overlap strongly with ISO 42001 Annex A:
- AI governance (CC1 / CC3) ↔ A.2, A.3, A.5 — policy, named owner, impact assessment.
- Model monitoring (CC7.2) ↔ A.6.2.6 — drift and quality monitoring.
- AI output transparency (CC2 / C1) ↔ A.8 — user disclosure and labelling.
- Training data (C1 / P4) ↔ A.7 — dataset data-sheets, consent basis.
The remaining three (prompt manipulation, data poisoning, automated decision-making) sit inside ISO 42001 but under different Annex A subsections. Points of focus are not mandatory, so this overlap narrows but does not eliminate the AI-only Annex A work.
Timing matters: the 2027 TSC is effective for engagements beginning on or after 31 December 2027. If you're building to ISO 42001 today, your controls already satisfy most of what the 2027 SOC 2 will require — no rework.
Frequently asked questions
How much of SOC 2 maps to ISO 42001?
Based on LowerPlane's practitioner mapping: roughly 40–50% of SOC 2 Type 2 evidence transfers to ISO 42001. The management-system scaffolding (clauses 4–10) reuses governance from your ISMS or SOC 2 programme, and information-security controls (access, change, vendor, monitoring) carry over. The 38 Annex A controls that are AI-specific (impact assessment, lifecycle, data, transparency) have no SOC 2 analogue and are net-new work.
Can one audit cover both SOC 2 and ISO 42001?
Not literally. SOC 2 is a US AICPA attestation; ISO 42001 is an international management-system certification issued by an accredited certification body. They are run by different audit firms (licensed CPA vs IAF-accredited CB). What you can combine is the evidence — one controls base, two reports — and an ISO 27001 + ISO 42001 integrated audit, which is standard CB practice.
Which ISO 42001 controls are new (no SOC 2 analogue)?
Annex A controls under A.5 (AI System Impact Assessment), A.6 (AI System Life Cycle), A.7 (Data for AI systems), A.8 (Information for interested parties), and A.9 (Use of AI systems). These cover impact assessment, model lifecycle stages, dataset documentation, transparency obligations and use-case governance — concepts SOC 2 does not touch directly.
Does the 2027 SOC 2 revision close the gap?
Partly. The AICPA's 2027 exposure draft adds 7 AI points of focus — AI governance, model monitoring, prompt manipulation, data poisoning, AI output transparency, training data, automated decision-making. Four of those (AI governance, model monitoring, AI output transparency, training data) overlap strongly with ISO 42001 Annex A. Points of focus are not mandatory controls, so overlap doesn't eliminate the AI-only Annex A work — but it reduces it.
Download the crosswalk, or turn it on in LowerPlane
LowerPlane's AIMS maps every SOC 2 control to its ISO 42001 clause and Annex A anchor, so evidence collected once reports against both.