Yes for the preparation, no for the audit. Claude or ChatGPT can do much of the SOC 2 preparation work with you: explain the criteria, question your practice, draft policies and the system description, and structure evidence. Neither can decide for your management, operate a control or attest. The SOC 2 report and its opinion come only from a licensed CPA firm that examines your evidence.
Key takeaways
- Claude and ChatGPT are tools for drafting and questioning. They shorten the writing. The decisions stay with your management.
- Three limits apply to any assistant: it cannot attest, it cannot see your systems unless given access, and it produces documents, not operating controls.
- A person at your company reviews everything an assistant writes before it is approved or sent to the auditor.
- The model to follow is AI-guided preparation, CPA-performed audit.
What can Claude or ChatGPT do for SOC 2?
An assistant is useful wherever the work is reading, asking or writing.
- Explain the criteria. The Security criteria are the nine common criteria series, CC1 to CC9, with 33 criteria in all. An assistant can restate each one in plain words. Check its answer against the AICPA text or our controls list by Trust Services Criteria.
- Interview you. Told to ask before it writes, an assistant will question how access is granted, how changes are reviewed and how leavers are removed. That is the start of a gap assessment.
- Draft documents. Policies, procedures and sections of the system description, written from your answers. See how to write SOC 2 policies with AI.
- Review a draft. Given a policy and your notes on real practice, it can list the sentences that the practice does not support.
- Help with exports. A coding agent such as Claude Code or Cursor can draft a read-only query for a population. A person reviews it before it runs. See evidence collection with an AI agent.
The whole workflow, from scope to report, is in SOC 2 with AI.
Which SOC 2 tasks can AI draft, and which need a person or the auditor?
Each task falls into one of three groups. “AI draft” means the assistant writes a first version and a person checks it. “Human decision” means management decides or a named person acts. “Auditor only” means the work belongs to the independent CPA firm.
| SOC 2 task | Group | What that means |
|---|---|---|
| Explain a criterion in plain words | AI draft | Check it against the AICPA text |
| First version of a policy or procedure | AI draft | Written from your answers; a person edits it |
| Sections of the system description | AI draft | Management verifies every statement |
| List of candidate risks | AI draft | Management scores them and adds its own |
| Read-only export query | AI draft | A person reviews it before it runs |
| Scope: system, categories, report type | Human decision | Management decides |
| Which controls to adopt | Human decision | Management designs its own controls |
| Risk treatment and acceptance | Human decision | A named person in management decides |
| Approval of a policy | Human decision | The owner approves and dates it |
| Access reviews, change approvals, offboarding | Human decision | People perform the control and keep the record |
| Management’s assertion | Human decision | Management provides it to the auditor |
| Tests of controls | Auditor only | Inquiry, observation, inspection and reperformance |
| Opinion and report | Auditor only | Issued by a licensed CPA firm |
How the auditor tests is explained in tests of controls, IPE and IUC.
What can Claude or ChatGPT not do in a SOC 2 audit?
AI cannot attest
SOC 2 is an attestation report with an opinion issued by a licensed CPA firm, under SSAE 18, AT-C sections 105 and 205. Software never forms the opinion. An assistant’s statement that you meet a criterion has no standing with a reader of the report. See who can perform a SOC 2 audit. For a company with no compliance platform, the examination is our Zero-Platform SOC 2 audit.
AI cannot see your systems unless given access
A chat assistant knows what you type or paste, and nothing else. Where it lacks a fact, it writes something plausible. An agent with read-only credentials can query a system, and from then on it is one more account under your access rules. Before you paste internal documents, read is it safe to put security documents into ChatGPT or Claude?
AI produces documents, not operating controls
A policy that says access is reviewed every quarter is not an access review. A Type I report covers the design of controls at a point in time. A Type II report covers design and operating effectiveness over a period, and the evidence is the record of people doing the work. AI-written policies and descriptions are acceptable only if they describe what your company actually does. The audit tests how controls operate, not how documents read.
A policy that describes more than you do leads to exceptions. See SOC 2 common findings and how to fix them.
What should we ask Claude or ChatGPT first?
Start with two prompts. Both make the assistant ask before it writes.
Prompt: explain a criteria series and question our practice
Act as a tutor for the AICPA 2017 Trust Services Criteria, common criteria CC1 to CC9. I am preparing my company for a SOC 2 examination against the Security criteria. Start with CC6, Logical and Physical Access Controls. Explain in plain words what the series asks of a company. Then ask me how we handle each topic today. Ask me one question at a time and wait for my answer. When I have answered, summarize what we do, what is missing and what you could not tell. Do not describe a control as in place unless I said so. Mark each assumption as an assumption. List anything you assumed that I need to verify.
Prompt: review a draft policy against what we do
Below are our draft access control policy and my notes on how access is granted, changed, reviewed and removed today. [paste the policy] [paste the notes] Ask me about anything in the notes that is unclear before you start. Then quote every sentence in the policy that the notes do not support, and say what the notes show instead. Suggest wording that matches the practice. Do not add requirements we do not follow. Flag each assumption you make about our practice. List anything you assumed that I need to verify.
Who does what
| Task | Your team (with or without AI) | Your AI assistant | The auditor |
|---|---|---|---|
| Learning the criteria | Reads the criteria; checks the assistant’s explanation | Explains each criterion and answers questions | Explains the criteria; sends the evidence request list once engaged |
| Drafting policies and the description | Supplies the facts; edits, verifies and approves | Drafts from your answers; lists its assumptions | Reads the documents and compares them with the evidence |
| Scope, controls and risk decisions | Decides; management owns each decision | Lists options and questions to consider | No role in management decisions |
| Operating controls | Performs each control and keeps the record | Cannot operate a control | Tests whether the controls operated, for a Type II report |
| Opinion and report | Provides management’s assertion | No part | Forms the opinion and issues the report |
Auditsuisse does not draft a client’s policies, configure its systems, choose its controls or make management decisions. Hands-on help comes from a separate provider.
Frequently asked questions
Can ChatGPT or Claude write our SOC 2 report?
No. The report is issued by a licensed CPA firm and carries the auditor’s opinion. An assistant can draft the system description, which is management’s part of the report, and management verifies it. See write your SOC 2 system description with AI.
Will the auditor accept policies written with Claude or ChatGPT?
On two conditions: the policy describes what you do, and a person at your company has reviewed and approved it. We read the policy and compare it with the evidence. What auditors accept from AI sets out our rules.
Is Claude or ChatGPT better for SOC 2?
We do not rank them. Either can explain criteria and draft documents. The result depends on the facts you supply and the review you do afterwards.
Do we still need a compliance platform?
No. A platform is one way to collect evidence, not a requirement. SOC 2 without Vanta or Drata lists what you need instead, and the Zero-Platform SOC 2 audit is our audit for companies without one.
Sources & further reading
AICPA, 2017 Trust Services Criteria (with revised points of focus, 2022); SSAE No. 18, AT-C sections 105 and 205; AICPA, SOC suite of services.
Related: SOC 2 readiness checklist, SOC 2 Type 1 vs Type 2, Audit methodology.
Product names are trademarks of their owners. Auditsuisse is not affiliated with these vendors.
Work with a licensed CPA firm
Auditsuisse Assurance NA PC is a licensed US CPA firm, enrolled in the AICPA Peer Review Program and authorized to perform SOC 2 examinations and issue SOC 2 reports. You prepare; we examine your evidence. See the Zero-Platform SOC 2 audit or book a scoping call.