United States · AI companies
SOC 2 for US AI Companies: Pass the Enterprise Security Review
What US enterprise, financial and federal buyers require from an AI vendor's SOC 2 report: boundary, controls, pitfalls and the CPA firm examination.
Did you know SOC 2 can cost less than US$10,000, all-inclusive? If you've been quoted more, talk to us before you sign.
100% success, on time: the Sterling Promise
We stay with you until your agreed outcome is achieved, through consulting, audit preparation and the resolution of audit findings. The engagement defines the standard, scope, outcome and responsibilities; your organization puts the agreed controls into use; a US-licensed CPA firm issues the SOC 2 report; an accredited certification body issues the certificate, and that decision stays independent.
Done for you, end to end
We write the policies, set up the controls and evidence, run the internal audit and coordinate the independent audit. You deal with one team from first call to final report.
Simple and quick
A free gap assessment, then one plan with dates from day one. Effivity, our GRC automation platform, is included to keep the evidence moving.
8,100+ client engagements
In 60+ countries over 20+ years, with experts who bring 30+ years of experience.
Under US$10,000, all-inclusive
For most start-ups and SMEs: SOC 2, ISO 27001, ISO 42001, HIPAA or ISO 9001, including the independent audit. A fixed quote follows a free gap assessment.
Included in every engagement
Every Sterling consulting engagement comes with the following, so you are not buying extras to get to the finish line:
- Effivity, Sterling's GRC automation platform, with a full-year subscription, to run your controls, evidence, tasks and audits in one place.
- Sterling Academy: preferential course access, with complimentary awareness training for your people.
- Internal auditor training for your audit team, as applicable to your standard.
- For information-security engagements, a vulnerability assessment.
These are part of the engagement, not a separate purchase. The details depend on the standard you are working toward, which we confirm during the free gap assessment.
Consulting and auditing are always separated: the people who consult never audit you. That keeps your internal audit objective and protects the independence of the CPA firm, Qualified Security Assessor or certification body that performs the final step.
Behind this are 8,100+ client engagements in 60+ countries over 20+ years, led by experts with 30+ years of experience.
All-inclusive pricing for start-ups and SMEs
Did you know SOC 2 or ISO 27001 can cost less than US$10,000, all-inclusive? If you've been quoted more, talk to us before you sign.
Sterling's all-inclusive fee for SOC 2, ISO 27001, ISO 42001, HIPAA or ISO 9001 is under US$10,000 for most start-ups and SMEs. It covers our done-for-you consulting and the independent audit: the SOC 2 report from a US-licensed CPA firm, as AICPA standards require, or the certification audit by an accredited certification body. For HIPAA, it covers HIPAA compliance consulting and an independent HIPAA assessment. Effivity, training and a vulnerability assessment are included. That's less than a month's cost of an in-house security hire: the US Bureau of Labor Statistics puts the national median annual wage for information security analysts at $129,180 (May 2025), about $10,765 a month in pay alone[1].
A fixed quote follows a free gap assessment.
What the fee covers, and what it doesn't:
- It is the fee for your first SOC 2 report or your first certification only.
- Annual work is always additional and billed the following year: later SOC 2 report periods, and ISO surveillance and recertification audits.
- It assumes a typical start-up or SME scope: one product or service and one main location.
- Larger or multi-site scopes get a fixed quote after the gap assessment.
- PCI DSS is quoted separately: a fixed quote follows a free gap assessment.
[1] Occupational Employment and Wage Statistics, May 2025: 15-1212 Information Security Analysts, national annual median wage. US Bureau of Labor Statistics. accessed October 23, 2026.
Why US enterprise buyers want a SOC 2 report from AI vendors
The outcome. A SOC 2 report your US buyers' security teams accept, so a pilot turns into a signed production contract. We prepare everything; the SOC 2 report is issued by a US-licensed CPA firm, as AICPA standards require.
SOC is a suite of service offerings CPAs may provide in connection with the system-level controls of a service organization[1]. US security reviewers know the format well, which is why an AI vendor meets the request so early. Four US moments trigger it:
- Pilot to production. The customer's third-party risk team steps in before prompts, documents or call recordings flow through your API.
- Data-sensitive buyers. In Sterling's experience, banks, hospital systems and law firms decline to send content to an AI vendor without a report.
- Cloud marketplaces and resellers. In Sterling's experience, listing reviews often require security evidence first.
- US investors. Diligence asks how customer data moving through the model is protected.
A SOC 2 report covers the controls around your service. It does not judge model accuracy or fairness; that is governance evidence, covered on our ISO/IEC 42001 page.
The US rules behind your buyer's questionnaire
The outcome. You answer regulated US buyers knowing what they must check, with evidence mapped to each duty.
- Lenders and fintechs. The FTC Safeguards Rule requires covered financial institutions to select capable service providers, require safeguards by contract and periodically assess providers based on risk[3].
- Public companies. SEC Regulation S-K Item 106 has registrants describe whether they have processes to oversee and identify cybersecurity risks associated with their use of third-party service providers[4].
- Federal agencies. OMB Memorandum M-25-22 (April 3, 2025) requires agencies to ensure contracts permanently prohibit using non-public inputted agency data and outputted results to further train publicly or commercially available AI algorithms, absent explicit agency consent[5].
None of these rules names SOC 2. A report is simply the evidence most US vendor teams already know how to read.
What a US AI vendor's SOC 2 system boundary covers
The outcome. A system description that matches what your buyer is purchasing, with no surprises in the examination.
- Inference path. The API or interface, routing, retrieval and orchestration layers, and model serving.
- Customer content. Where prompts, files, embeddings and outputs are stored, for how long, who sees them and whether they improve models. US reviewers read this paragraph first.
- Training and evaluation. In scope where customer data or production access reaches it.
- Model providers. An external foundation model is a vendor in your system, and your oversight of it is tested.
- ML tooling. Notebooks, feature stores, labeling tools and model registries.
The AICPA lists five trust services categories: security, availability, processing integrity, confidentiality and privacy[2]. AI vendors commonly add confidentiality, and availability when the API sits inside customer workflows. We word your system commitments so you never promise model accuracy the report cannot support.
Controls US security reviewers probe at AI companies
The outcome. Clean answers to the questions that stall AI deals.
| Area | What US reviewers check | What we put in place with you |
|---|---|---|
| Customer content access | Can engineers read prompts and outputs? | Role-based access, approved break-glass viewing, logged and reviewed |
| Retention and deletion | How long is content kept, and can it be erased? | A schedule the product and storage layers enforce |
| Model and data suppliers | Which outside models and datasets do you rely on? | A vendor register with due diligence, contract terms and change watch |
| Model releases | How does a new model version reach production? | Evaluation results, an approver and a rollback path per release |
| Secrets and weights | Where do API keys and weights live? | Managed secrets, rotation and restricted registry access |
| Misuse and leakage | How would you detect data leaking through the product? | Alerts, an incident process and a record of a drill |
Where US AI companies lose time on SOC 2
The outcome. We clear these before the CPA firm starts fieldwork.
- Promising accuracy. Describe how data is handled and protected, not how good the outputs are.
- Production credentials in notebooks. A favorite exception.
- Prompts logged to unknown tools. Know every place customer content lands.
- The model provider left out. Reviewers want your oversight of it on record.
- Training on customer data without clear terms. Contract, settings and system description must say the same thing.
- A hiring surge mid-observation. We time the Type 2 window around your hiring and release calendar.
- Chasing shortcuts. The AICPA's SOC page features a February 2026 Journal of Accountancy article, "Promises of 'fast and easy' threaten SOC credibility"[1]. We make you quick by preparing well, not by cutting corners.
SOC 2 first, AI governance next (Sterling's recommendation)
The outcome. Each framework you add reuses the records from the last.
US security teams ask about data protection early in a deal; AI and legal reviewers arrive later. A SOC 2 report therefore tends to unlock the first contracts, and ISO/IEC 42001 helps when the same buyers deepen the review. Access reviews, change approvals, vendor assessments and incident drills carry across, kept once in Effivity. If a contract already demands AI governance first, we reverse the order.
How we take US AI companies to a SOC 2 report
The outcome. A report on schedule, done for you, with one point of contact. Sterling's record is 100% success and on-time completion across 8,100+ client engagements in 60+ countries over 20+ years, led by experts with 30+ years of experience.
1. Free gap assessment. Inference path, data handling, vendors and existing controls, with a proposed boundary and categories. 2. Policies and controls. Designed around how your ML and platform teams already work. 3. Evidence. Collected in Effivity, Sterling's GRC automation platform. 4. Readiness review. We test as an examiner would, and you close the gaps. 5. Examination. The SOC 2 report is issued by a US-licensed CPA firm, as AICPA standards require. Sterling coordinates scheduling, evidence requests and walkthroughs, and the people who consult never audit you.
Want the report in hand before your next renewal? Get a proposal and we will set your dated plan.
How Sterling delivers SOC 2
A SOC 2 engagement has one deliverable that no consultant can hand you: the report. The SOC 2 report is issued by a US-licensed CPA firm, as AICPA standards require[1]. What Sterling does is make sure you get there ready, and that nothing between today and the report lands on your desk.
With Sterling, you deal with one team from start to finish. We coordinate:
- a readiness review and a gap assessment against the Trust Services Criteria;
- the policies, procedures and controls you need, written for how your company actually works;
- evidence collection, so the examination does not become a scramble;
- an internal audit before the examiner arrives, carried out by people who did not advise you on the controls;
- the Type 1 or Type 2 path that fits your deal timeline, and the examination itself, including scheduling, requests and follow-ups.
Our experts bring 30+ years of experience, and the firm has completed 8,100+ client engagements in 60+ countries over 20+ years. You get a single plan, a single point of contact and a clear view of what happens next.
[1] SOC (System and Organization Controls) Suite of Services. AICPA & CIMA. accessed October 23, 2026.
Sources
1. System and Organization Controls: SOC Suite of Services (incl. listed article 'Promises of fast and easy threaten SOC credibility', Journal of Accountancy, Feb 1, 2026). AICPA & CIMA. accessed October 26, 2026.
2. 2017 Trust Services Criteria (With Revised Points of Focus - 2022). AICPA & CIMA. accessed October 23, 2026.
3. 16 CFR 314.4 Elements (Standards for Safeguarding Customer Information), paragraph (f). Electronic Code of Federal Regulations (FTC Safeguards Rule). accessed October 26, 2026.
4. 17 CFR 229.106 (Item 106 of Regulation S-K) Cybersecurity, paragraph (b)(1)(iii). Electronic Code of Federal Regulations (SEC). accessed October 26, 2026.
5. OMB Memorandum M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government (April 3, 2025). Executive Office of the President, OMB (whitehouse.gov). accessed October 26, 2026.
Information only
The information on this page is provided for general informational purposes only. It is not legal advice, and it is not a substitute for advice from a qualified attorney about your own situation.
Requirements depend on your scope, your customers' contracts, your industry and the regulations that apply to you. Two companies asking for the same standard can have very different obligations. We confirm yours during a gap assessment, and we update our content as standards change, but you should check the current text of any standard or rule that matters to you.
Sterling consults and prepares. It does not decide whether you meet a standard. That decision belongs to the independent party that performs the final step: the US-licensed CPA firm for a SOC 2 report, the Qualified Security Assessor or acquirer for PCI DSS, and the accredited certification body for ISO certificates. Timelines and outcomes depend on your readiness and on those parties.
Standards and marks named on this site belong to their respective owners.
FAQs
Frequently asked questions
Is SOC 2 an AI governance standard?
No. A SOC 2 examination looks at the controls around a service system, such as access, change management, monitoring and vendor oversight. It does not evaluate model accuracy, fairness or how training data is governed. Many US AI companies use the report as their baseline security proof for enterprise deals, then add an AI management system such as ISO/IEC 42001 when buyers start asking governance questions about the models themselves.
Who issues the SOC 2 report for an AI company?
The SOC 2 report is issued by a US-licensed CPA firm, as AICPA standards require[1]. Sterling prepares your controls, runs a readiness review that tests them the way an examiner would, and coordinates the whole examination, from scheduling to evidence requests and walkthroughs. You deal with Sterling end to end, and the people who consult never audit you.
Which trust services categories do AI vendors usually choose?
Security is always the base. Of the five categories the AICPA recognizes[2], confidentiality is the most common addition for AI vendors, because customer content is the asset buyers care about. Availability is added when your API sits inside a customer's live workflow. We choose with you based on your contracts and buyer questions, since each extra category adds testing and evidence.
Do our prompts and outputs count as customer data?
For most AI products, yes. The system description should state where prompts, files and outputs are stored, for how long, who can access them, and whether any of it is used to improve models. US reviewers read that section closely, so the description, your contract terms and the product's settings must agree. Mismatches between them are a frequent cause of stalled reviews.
Does calling an external model API affect our report?
It affects your vendor management controls. The model provider becomes a supplier in your system, so you need to show how you evaluated it, what the contract says about your customers' data, and how you learn about and respond to model changes. We build that vendor register with you and keep the evidence current in Effivity.
Should a US AI start-up do Type 1 or Type 2 first?
Our recommendation for most US start-ups is to plan for Type 2, since enterprise reviewers eventually require it, and to take a Type 1 first only when a specific deal needs a report sooner. A Type 1 looks at control design at a point in time; a Type 2 also covers operation over a period. We set the timing after the gap assessment, around your sales calendar.
Can the research team keep its notebooks and experiments?
Yes, with sensible boundaries. Experimental work continues as long as production data and credentials are controlled, access is justified and approved, and environments are described accurately in the system description. We design guardrails that researchers can work with, so security does not slow model development down. In practice that means separate credentials for research, no copies of production customer content without approval, and a short record of any exception.
What is included when a US AI company works with Sterling?
Every engagement includes a full-year subscription to Effivity, our GRC automation platform; preferential Sterling Academy course access, with complimentary awareness training for your people; internal auditor training for your audit team, as applicable to your standard; and, for information-security engagements, a vulnerability assessment. The CPA firm's examination sits inside the all-inclusive fee.
Keep exploring
More in Compliance for AI companies in the US
Compliance for AI companies in the US
Free gap assessment
Start with a free gap assessment
Not sure where you stand? Start with a free gap assessment. In one conversation, a Sterling consultant looks at what your customers, investors or contracts are asking for, compares it with what you have in place today, and tells you what is missing, in what order to fix it and how long it is likely to take.
usa@sterling-grc.com+1 201 263 4443 (global headquarters) · A Sterling consultant replies within one working day
What would you like to achieve?
Choose a starting point. The enquiry form asks for your name, work email, phone with country code and company.
Book my free gap assessmentContinue to the enquiry form to share your details. Please do not include passwords, sensitive personal information or confidential files. A Sterling consultant replies within one working day. Privacy notice