SOC 2 for AI companies: what a model provider changes
An AI product sits for the same SOC 2 as any other software. The model provider moves which criteria get tested hard, and what proves them.
SOC 2 for AI companies is the same examination any software company takes. The AICPA criteria have no section on models.1 What shifts is the weight on a handful of existing criteria once customer data leaves your stack for a model provider.
That provider becomes a subservice organization, your vendor file becomes the most read part of your evidence, and your data processing agreement ends up in front of buyers who read it line by line. None of it needs an AI specialist.
Start with scope, because it settles the rest
An auditor tests the controls in your system description against the criteria in your scope.2 For a first report that scope is Security. It is the category a buyer means when they ask if you have SOC 2, and it is the only category we scope. The reasoning is laid out in the scoping tool.
Security does not test your model. Output quality, bias, hallucination rates and eval coverage are real concerns, but they are not security criteria, so no auditor will sample them. Security asks narrower questions: who can reach the data, where it goes when it leaves, how a change gets approved, and whether anyone noticed when something broke.
Governance of the AI itself has its own standard. ISO/IEC 42001 was published in December 2023 and describes a management system for artificial intelligence. An accredited certification body certifies it. A SOC 2 will not stand in for that certificate, and the certificate will not stand in for a SOC 2 when a buyer asks for one.
Six criteria a model provider puts under pressure
Every inference call that carries customer data crosses your system boundary. That single fact wakes up six common criteria.1 Here is why each one matters and what to have on file before fieldwork.
| Criterion | Why the provider matters here | Evidence to have ready |
|---|---|---|
| CC3.2 risk identification | A new exit route for customer data, which needs an owner on the risk register | A risk assessment dated inside the period that names the provider and the data it gets |
| CC6.1 logical access | The API key is a production secret like any other | The secrets manager entry and a record of its last rotation |
| CC6.7 restricted transmission | Data leaves the boundary on every call, so this is the center of the question | The data flow in the system description, encryption in transit, and the rule for what may enter a prompt |
| CC7.1 and CC7.2 monitoring | Prompt logs are a log source and also a second copy of customer data | The retention setting on prompt logs and a query that returns your oldest claimed record |
| CC8.1 change management | A provider can ship a new model version without touching your repo | A pinned version in config plus the ticket and approval for each upgrade |
| CC9.2 vendor risk | The provider is a subservice organization that has to be disclosed | The vendor assessment, the signed agreement, and the provider report with a dated read note |
Look at the right column. None of it is exotic, because each row is a familiar control holding a new object, and it gets tested the familiar way: a population, a sample, a date and an artifact you can open. For the criteria in plain English, our Trust Services Criteria guide walks through all five categories.
Treat the provider as a subservice organization
A subservice organization is a vendor your own service commitments rely on.3 Ask two questions. Does your product stop doing what you promised when the provider is down, and would a mishandled prompt on their side break a promise you made to a customer? A yes to either question means the provider qualifies.
You can include it or carve it out. Inclusion means the provider joins your examination with its own assertion and its own testing, and a ten person startup will not get that from a frontier lab. So you carve out: name the provider, exclude its controls from the description, and state in writing what you assume it does.
Those assumptions are written as complementary subservice organization controls. Word them so a reader can see exactly what you rely on. Our page on wording CUECs and CSOCs shows a sentence that transfers a duty next to one that only fills space.
Three answers your vendor file needs
A model provider assessment turns on what happens to the text you send, not on the provider’s architecture. Record three answers, each tied to a document you can open.
- Training on your inputs
- Use the signed agreement or the data processing addendum it points to. A marketing page or a support reply is not evidence. Write the section number into the vendor file. If prompts carry customer data and the agreement says nothing about training, treat the gap as a finding and resolve it.
- Retention at the provider
- How long prompts are kept, in which region, and who on their side can read them. If you cannot state that period, you cannot give it to your own customer when they ask.
- Their subprocessors
- The provider has vendors too, and the one that counts is wherever inference physically runs. Region matters if you promised a customer data residency, so get the list and the notice terms for any change to it.
Keep all of it in the file. An auditor tests a dated assessment, the signed agreement, the provider report and a note naming who read it, and a link to a trust page is none of those. This is the same request that already appears on the auditor’s PBC list, and it reads the same for a model vendor as for a payroll vendor.
Pin the model version like any other dependency
Here is the one operational wrinkle that feels new. Your provider retires a version or pushes a point release, outputs shift, and nothing went through your change process on the way in. Behavior in production now has no approved change behind it, and that is what an auditor looks for when they pull your change population.
The fix is dull. Pin the version in configuration so an upgrade only happens when you choose it. Treat an upgrade like a schema migration, with a ticket, a reviewer and an approval before it reaches production. Then say so in your change management policy, because a policy that only covers your own deploys does not describe your real system. Our policy count page shows where that wording lives.
Decoding the AI section of a security review
Security reviews now carry a block of AI questions, and they are mostly about a vendor you do not control. Your report answers some of them and cannot answer the rest, so here is what each question is really after and the document that settles it.
| What they ask | What they want to know | What you send |
|---|---|---|
| Which third party models do you use? | If there is an undisclosed data path | The vendor register and the system description that names the carved-out provider |
| Is our data used for training? | The training clause, both theirs and yours | The section number in the signed agreement, plus your own policy on fine tuning |
| How long are prompts kept? | If your retention promise holds at the provider | The provider’s stated period and your own prompt log retention setting |
| Can we turn AI features off? | If the data path is optional | The product setting, plus a CUEC if the customer has to flip it |
| What keeps sensitive fields out of prompts? | A control, not a promise | The redaction or allow list in code and a test showing it runs |
| Does your SOC 2 cover the AI features? | Scope and timing | The system description and the report period, read side by side |
A report covers the system it describes, for the period it covers. A feature that shipped after the period closed is not in it, however good the feature is. Say that plainly on the form. A reviewer can check an overstatement against the report you sent them in about a minute.
What this costs, and what you can do yourself
None of the work above is AI specific, so it should not carry an AI surcharge. For context, here is what the outside routes cost according to the people who sell or track them.
- Comp AI states that a vCISO or compliance consultant might charge $150 to $400 an hour, which can total $20,000 to $50,000 for a full SOC 2 prep engagement. Source, checked 2026-07-30.
- Vendr reports a median annual contract value of $20,000 for Vanta, based on purchases completed through its marketplace. Source, checked 2026-07-30.
Much of this page is work a founder can do in an afternoon: pull the training clause, pin the model version, set prompt log retention, and add the provider to the risk register. Those cost time, not money. The part that needs paid help is the examination itself, and that has to be a licensed CPA firm. cybersoftware is not a CPA firm. SOC 2 examinations are performed by independent licensed U.S. CPA firms.
Our software is $199 a month, cancel any time, or $2,189 a year, pay for eleven months, get twelve. Audits go through our preferred pricing program, and we negotiate the fee on your behalf, and you see the price in your account before you book. A Type 2 is quoted after the 3-month observation window. Using a model provider changes none of those numbers. It only changes what goes in your vendor file.
Your next step
Find out where you stand before you spend anything. The free readiness assessment takes about 15 minutes and asks about vendors, access and change control, including the model provider. You get a score and a gap list you can work through yourself. When you are ready to compare plans, see pricing for the monthly and yearly options side by side.
Questions
Does an AI company need a different kind of SOC 2?
Should the model provider be carved out of my SOC 2?
Will the SOC 2 report show whether my model provider trains on customer data?
Is switching to a new model version a change for SOC 2 purposes?
How does ISO 42001 relate to SOC 2?
Sources
Get audit-ready without a compliance team
The readiness assessment is free, with no payment and no card. When you are ready, the software is $199 a month, cancel any time, and audits go through our preferred pricing program. You can be audit-ready starting at about a week.
Start with a free readiness assessmentcybersoftware is not a CPA firm. SOC 2 examinations are performed by independent licensed U.S. CPA firms.