Do AI startups need SOC 2 Type I or Type II?

Many start with a Type I to show control design quickly, then move to a Type II, which covers operating effectiveness over a period and is what most enterprise buyers ultimately want. Some go straight to Type II.

Is SOC 2 enough for an AI company?

It proves security, but it is not an AI-governance standard. As AI scrutiny and rules like the EU AI Act grow, many AI companies pair SOC 2 with ISO/IEC 42001 or the NIST AI Risk Management Framework. On Keel, shared controls carry across all of them.

Does Keel perform the SOC 2 audit?

No. Only a licensed CPA firm can issue a SOC 2 report. Keel is where you prepare for it: map your controls to the Trust Services Criteria, collect the evidence, and give your auditor a free seat to review it.

How do we answer AI security questionnaires before the report is ready?

From what you already have in place. On every paid plan, Keel drafts answers to a prospect’s questionnaire from your own controls and policies, cites the source for each one, and marks anything it cannot support for review. Your trust center can share approved policies and your subprocessor list while the audit is still in progress.

AI companies

SOC 2 for AI companies

For an AI company, SOC 2 is usually the first report enterprise buyers ask for before they will send you their data. It is an attestation, performed by a licensed CPA firm, against the AICPA Trust Services Criteria.

Start free SOC 2 overview

Context

Why it matters for AI companies

AI products handle sensitive customer data, and buyers know it, so security review is often stricter for AI vendors. A SOC 2 report (usually Type II, which covers a period rather than a single date) is the fastest way to answer those questions once and unblock revenue.

Priorities

What to focus on

01

Security is required; the rest follow your promises

SOC 2’s Trust Services Criteria are Security, Availability, Processing Integrity, Confidentiality, and Privacy. Security (the Common Criteria) is always required; add Confidentiality or Privacy if you make those commitments about customer or training data.

02

The AI-specific questions buyers ask

For AI vendors, reviewers focus on how you handle customer data in prompts and training, who and what can access models and data (access control and MFA), and which third parties (including model providers) act as subprocessors. Map those clearly and evidence them.

03

Pair it with AI governance

SOC 2 proves security posture but is not an AI-specific standard. Buyers increasingly also ask how you govern AI itself, so pairing SOC 2 with a framework like ISO/IEC 42001 or the NIST AI RMF is becoming the strong position. Keel lets you run them on one crosswalked control library.

04

Treat model providers as vendors

If customer data reaches a third-party model API, that provider is part of your system. SOC 2’s vendor and business partner criterion (CC9.2) applies to it as it does to your cloud host: diligence before onboarding, contract terms, and ongoing review. Keel’s AI vendor assessment template asks a provider whether customer data trains shared models, which models the service depends on, and whether data stays in agreed regions. Your trust center can list the provider as a subprocessor.

05

Run model and prompt changes through change management

Swapping a model version or editing a production system prompt changes how your product handles customer data. Putting those changes through the same request, test and approval path as code (CC8.1) gives the auditor a record to sample and gives you a way to trace a change in output back to a decision.

06

Keep a list of the AI you build and use

SOC 2 does not ask for an AI inventory by name, but buyers’ questions are easier to answer with one. Keel’s AI services register records each service’s vendor, purpose, the categories of data it processes, an owner, a risk tier and an approval status, and the same list later supports ISO/IEC 42001 or EU AI Act work.

Do it once, not twice

Keel is built on one crosswalked control library, so a control you implement for SOC 2 counts toward every other framework it satisfies. Add a second framework later and it mostly reuses this work. See the crosswalk explorer for the exact overlap.

Start free Check your readiness

Common questions

Do AI startups need SOC 2 Type I or Type II?: Many start with a Type I to show control design quickly, then move to a Type II, which covers operating effectiveness over a period and is what most enterprise buyers ultimately want. Some go straight to Type II.
Is SOC 2 enough for an AI company?: It proves security, but it is not an AI-governance standard. As AI scrutiny and rules like the EU AI Act grow, many AI companies pair SOC 2 with ISO/IEC 42001 or the NIST AI Risk Management Framework. On Keel, shared controls carry across all of them.
Does Keel perform the SOC 2 audit?: No. Only a licensed CPA firm can issue a SOC 2 report. Keel is where you prepare for it: map your controls to the Trust Services Criteria, collect the evidence, and give your auditor a free seat to review it.
How do we answer AI security questionnaires before the report is ready?: From what you already have in place. On every paid plan, Keel drafts answers to a prospect’s questionnaire from your own controls and policies, cites the source for each one, and marks anything it cannot support for review. Your trust center can share approved policies and your subprocessor list while the audit is still in progress.

More guides: SOC 2 on Keel · SOC 2 for SaaS companies · SOC 2 for fintech companies · SOC 2 for healthcare and healthtech · SOC 2 for education and edtech · ISO 27001 for AI companies · ISO 42001 for AI companies

Framework names are referenced factually for guidance. Keel is not affiliated with or endorsed by the bodies that publish them. See our legal and trademarks page.