AIUC-1 vs SOC 2 vs ISO 42001: Which Do AI Startups Need?

Through 2025 and 2026 three different buyer-side signals converged on AI startups. US enterprise security teams kept asking for SOC 2. EU buyers and large-financial-services procurement started asking for ISO 42001. And buyers whose core concern is agents — the ones who evaluate how your product takes action, not just how it generates text — started asking for AIUC-1. This piece lays out what each actually certifies, how they overlap, and the order a 1–50 person AI startup should pursue them in.

Side-by-side

Factor	AIUC-1	SOC 2	ISO 42001
What it certifies	Agent behaviour, tool use, oversight and incident response	Service-wide controls against AICPA Trust Services Criteria	AI management system (governance, lifecycle, risk)
Type	Certification	Attestation report	Certification
Issued by	Accredited AIUC-1 certification body	Licensed CPA firm	Accredited ISO certification body
Output	Public certificate	Report distributed under NDA	Public certificate
Duration	Typically annual recertification	Report valid ~12 months (bridge letter → next)	Certificate valid 3 years + annual surveillance
Timeline from cold start	2–4 months (varies by scope)	30–60 days Type 1; 3–12 months Type 2	4–6 months
Buyer signal	Agent-heavy enterprise and regulated-industry procurement	US enterprise default	EU buyers, large financial services, procurement with AI checklist
AI-specific coverage	Deep — agent, tool use, action	Via TQA 9561 interpretation of existing TSC	Deep — management system for AI

What each actually covers

SOC 2 is the broadest of the three in scope and the narrowest in AI-specificity. It covers how you run the service — access, change management, monitoring, availability, confidentiality, privacy — and treats AI as part of the service rather than a separate domain. AICPA TQA 9561 is the current authoritative guidance on how AI slots in; a TSC revision with AI-specific points of focus is expected in 2027.

ISO 42001 is the broadest in AI-specificity and management-system depth. It requires you to establish an AI management system — AI governance, AI policy, risk assessment, resources, operations, performance evaluation, continual improvement — with Annex A controls covering impact assessment, lifecycle, data for AI, information for interested parties, and third-party relationships. It is purpose-built for the organisation's AI practice.

AIUC-1 is narrowest in scope but deepest in agent coverage. It focuses on agent behaviour: how an agent is instructed, what tools it can call, how actions are logged, when humans must approve, how incidents are detected and responded to, and how the agent handles data. It's most useful when your product is itself an agent, or when your buyer's primary concern is autonomous action on their systems.

Who issues them, and what the audit looks like

SOC 2: issued by a licensed CPA firm registered with a US state board of accountancy and subject to AICPA peer review. The engagement follows SSAE 18, with planning, walkthroughs, control testing and (for Type 2) sampling across the examination period. See how to verify your SOC 2 auditor for the checks.
ISO 42001: issued by an accredited certification body (ANAB, UKAS, DAkkS, SCC and similar national accreditation bodies). The audit is Stage 1 (documentation review) and Stage 2 (implementation audit), followed by surveillance audits each year of the 3-year cycle.
AIUC-1: issued by an accredited AIUC-1 certification body. The audit model is maturing; expect a documentation review and an operational audit of the agent control set, with annual recertification. Confirm the current audit model with the specific certification body you're considering.

Buyer signals by deal type

Buyer type	Likely asks for
US SaaS enterprise	SOC 2 Type 2
US financial services or healthcare	SOC 2 Type 2 + ISO 42001 for AI features
EU enterprise	ISO 27001 + ISO 42001 (SOC 2 as supplement)
Agent-first buyer	AIUC-1 as a differentiator on top of SOC 2
UK public sector	ISO 27001 + Cyber Essentials; ISO 42001 for AI-adjacent contracts
Singapore / APAC	SOC 2 + ISO 27001; AI Verify adjacent to ISO 42001

Recommended order for a 1–50 person AI startup

Days 0–30: SOC 2 Type 1. Scope lock, policy pack, integrations, evidence collection, auditor kickoff, bridge letter. Signable report at the end of month one. See the 30-day Type 1 playbook.
Months 2–5: ISO 42001. Stand up the AI management system clauses on top of the SOC 2 foundation. Internal audit cycle at month 4, Stage 1 at month 5, Stage 2 at month 6. Reuses ~60% of the SOC 2 evidence.
Months 2–13: SOC 2 Type 2. Observation window starts the day the Type 1 is issued (or earlier). Type 2 report issued in month 13 covering a 12-month period.
Add AIUC-1 when driven by a specific deal. Scope narrowly to the agent surface area the deal cares about. Reuse the model inventory, tool-scope register, action log and incident runbook you already built for SOC 2 + ISO 42001.

Shared evidence that feeds all three

AI/model inventory with model, provider, data flows, risk category.
Tool-scope register per agent (see SOC 2 controls for AI agents and MCP).
Risk register with AI-specific entries.
Vendor file for every model provider and third-party MCP server.
Change log for prompts, tool definitions and model versions.
Action log schema with user, agent, tool, outcome and approver.
Incident runbook covering prompt injection, data leakage and model outage.
AI use policy and acceptable use policy.
Human oversight model per feature (none, confirmation, explicit approval).

Frequently asked questions

What is AIUC-1?

AIUC-1 is a certification focused on AI agents and autonomous systems that emerged in 2025–2026 as enterprise buyers started asking for AI-specific assurance beyond ISO 42001. It covers agent behaviour, tool use, data handling, oversight and incident response. The certification body model and public register are still maturing — check the current status with the issuing body before you commit.

Is AIUC-1 a replacement for SOC 2?

No. SOC 2 is an attestation on service-wide controls under AICPA Trust Services Criteria; AIUC-1 is a certification of agent-level controls. Enterprise buyers who ask for AIUC-1 still ask for SOC 2. Think of AIUC-1 as a supplement specific to agent risk, not a replacement for the broader security report.

Which should I do first as an AI startup?

For US-first AI startups selling to enterprise, SOC 2 Type 1 first (30 days), then ISO 42001 (4–6 months), then AIUC-1 when a buyer specifically asks for it. For agent-first products where buyers are leading with AI governance questions, start ISO 42001 and SOC 2 in parallel and add AIUC-1 as a differentiator.

Does LowerPlane support AIUC-1?

LowerPlane maps shared evidence — risk register, model inventory, tool-scope register, incident response, vendor review — to the AIUC-1 control areas, SOC 2 Trust Services Criteria and ISO 42001 clauses so one evidence base feeds all three. Confirm the current scope of AIUC-1 support with our team if the certification is a hard contractual requirement.

How much do these frameworks overlap?

The three share about 40–55% of operational evidence (risk, change, access, incident, vendor, monitoring). AIUC-1 adds agent-specific controls (tool scopes, action logs, approval gates) that SOC 2 and ISO 42001 don't fully cover. ISO 42001 adds the AI management system layer (lifecycle, impact assessment, data quality). SOC 2 adds the service-level system description and subservice organisation treatment.

Are AIUC-1 and ISO 42001 both certifications?

Yes — both result in a certificate issued by a third party, both have surveillance cycles, and both are public signals you can show on your website. SOC 2 is different: it is a confidential attestation report distributed under NDA.

AIUC-1 vs SOC 2 vs ISO 42001 for AI agent startups