Key takeaways
- The AICPA published a draft of new SOC 2 Trust Services Criteria at the end of October 2026. Comments are open until January 15, 2027; if the draft is approved, it applies to engagements beginning on or after December 31, 2027.
- For engagements that begin in 2027, the 2017 Trust Services Criteria with the points of focus revised in 2022 will generally still apply. You can switch earlier only after final approval, and only if you agree on early adoption with your service auditor.
- The five categories (security, availability, processing integrity, confidentiality, and privacy) stay. Below them, the AICPA consolidates, removes, adds, and rewrites criteria, so control matrices tied to numbers such as CC6.1 will need remapping.
- New points of focus cover seven AI topics: AI governance, model monitoring, prompt manipulation, data poisoning, transparency of AI-generated outputs, training data management, and the risks of automated decision-making.
- If you already handle AI risk under ISO 42001, ISO 27090, or the EU AI Act, you can reuse that work for SOC 2. The outcome is still a SOC 2 report from a CPA firm, not a certification.
The SOC 2 Trust Services Criteria are the benchmark a CPA firm uses to evaluate your controls in a SOC 2 examination. The criteria themselves have not changed since 2017; in 2022, the AICPA revised only the explanatory points of focus. The draft for 2027 is the first fundamental revision of the criteria since 2017, and it is now open for comment.
If your company runs an ISMS under ISO 27001 and needs a SOC 2 report because US customers ask for it, the question that matters is not what changes in the US. It is which of your existing controls, evidence, and AI measures still hold. Our overview of SOC 2 as a framework explains what SOC 2 examines and how a report is structured.
What the AICPA plans for 2027: timeline and goals
The AICPA's Assurance Services Executive Committee (ASEC) issued the draft, and its SOC 2 Working Group accepts comments until January 15, 2027. The draft appeared at the end of October 2026: the AICPA download page is dated October 26, and the AICPA's press release is dated October 28, 2026.
The AICPA names two goals. SOC 2 reports should become clearer and more usable for the people who read them. At the same time, the criteria should stay relevant as AI, automation, and other emerging technologies change third-party risk.
| Step | Date | Status as of October 29, 2026 |
|---|---|---|
| Draft published | End of October 2026 (press release dated October 28, 2026) | Done |
| Comment period closes | January 15, 2027 | Announced |
| Final criteria | 2027, exact date not announced | Depends on approval after the comment period |
| Effective for | Engagements beginning on or after December 31, 2027 | Subject to approval; early adoption permitted |
A common misconception: the new criteria already apply to your 2027 SOC 2 examination. For engagements that begin in 2027, the 2017 Trust Services Criteria with the 2022 revised points of focus will generally still apply. You can switch earlier only once the final criteria exist and you agree on early adoption with your service auditor.
This matters most for Type 2 reports. The AICPA ties the effective date to when the engagement begins, not to when the reporting period ends. Ask your service auditor early how they will treat a review period that runs past December 31, 2027.
Alongside the Trust Services Criteria, the AICPA download page also announces a draft of the Description Criteria, which management uses to prepare the system description in a SOC 2 report. As of October 29, 2026, however, the page offers only two files, both for the Trust Services Criteria: a clean version and a redlined version.
What changes in the SOC 2 Trust Services Criteria
The five categories stay: security, availability, processing integrity, confidentiality, and privacy. Today, security is the only mandatory category, and its common criteria apply in every SOC 2 examination. The 2017 criteria organize them into 33 criteria in nine series, from CC1 (control environment) to CC9 (risk mitigation).
What changes is the structure underneath. According to the AICPA, the draft makes four kinds of changes: it consolidates duplicative concepts, removes criteria where appropriate, adds criteria where greater clarity or emphasis is needed, and rewrites existing language so each criterion is concise and uniformly understood. The AICPA offers the criteria themselves as a draft download after free registration.
| Aspect | 2017 Trust Services Criteria (current) | 2027 draft |
|---|---|---|
| Official title | 2017 Trust Services Criteria, with revised points of focus from 2022 | Proposed 2027 SOC 2 Trust Services Criteria |
| Status | In use | Draft, comments due January 15, 2027 |
| Applies to | Engagements until superseded | Engagements beginning on or after December 31, 2027, early adoption permitted |
| Categories | Security, availability, processing integrity, confidentiality, privacy | The same five categories |
| Common criteria | 33 criteria in nine series, CC1 to CC9 | Consolidated, partly removed, partly added, rewritten for consistency; the AICPA's announcement gives no new total |
| AI | The 2022 points of focus address emerging technologies and threats in general | New points of focus on seven explicitly named AI topics |
For your control matrix, this means the numbering will not stay stable. When criteria are consolidated, removed, and added, a control that sits under CC6.1 today may move to a different number, and CC6.1 itself may describe a different requirement. If you mapped controls one to one to criterion numbers, you will need to remap them.
The mistake many teams make right now is rebuilding the control matrix around the draft. The text can still change after the comment period, and your next examination will use the 2017 criteria anyway. A better move is to add a mapping column that you fill in once the final criteria are out. Our mapping of ISO 27001, GDPR, and SOC 2 shows how ISO 27001 Annex A maps to the common criteria today; it follows the 2017 numbering.
AI in SOC 2: what the new points of focus mean
AI enters the SOC 2 examination through new points of focus. The AICPA gives AI governance, model monitoring, prompt manipulation, data poisoning, transparency of AI-generated outputs, training data management, and the risks of automated decision-making as examples. In its announcement, the AICPA mentions AI explicitly only in connection with these points of focus.
Points of focus describe what matters for a criterion. They are not a mandatory checklist: the 2017 criteria state that you do not have to assess whether each point of focus is addressed. That does not mean, as many teams conclude, that AI plays no role in a SOC 2 examination. Points of focus steer what auditors ask. If a system in scope contains AI components, the CPA firm will ask how the relevant criteria are met for those components.
Prompt manipulation and data poisoning are not new topics for an ISMS. Our article on prompt injection in an ISO 27001 ISMS shows how to assess it in the risk register and which Annex A controls apply. The table below maps the seven topics to the frameworks where you probably handle them already. The auditor questions are our assessment, not AICPA wording.
| AI topic in the draft | Likely auditor question (our assessment) | Where you probably cover it already |
|---|---|---|
| AI governance | Who decides on new AI use cases, and is there an AI policy with clear roles? | ISO 42001, Annex A.2 and A.3; EU AI Act Art. 4 (AI literacy) |
| Model monitoring | How do you detect that model outputs degrade or become anomalous? | ISO 42001, Annex A.6; ISO 27001 A.8.16 (monitoring activities) |
| Prompt manipulation | Which inputs does the system process that you do not control, and what may it do with the result? | ISO 27090, clause 6; ISO 27001 A.8.26 and A.8.28 |
| Data poisoning | Who may change training or fine-tuning data, and how is that checked? | ISO 27090, clause 6; ISO 27001 A.5.21 and A.8.25 |
| Transparency of AI-generated outputs | Can users tell that a response or piece of content came from AI? | EU AI Act Art. 50 (since August 2, 2026); ISO 42001, Annex A.8 |
| Training data management | Where does the training data come from, who owns its quality, and how long is it kept? | ISO 42001, Annex A.7; GDPR Art. 6 for personal data |
| Automated decision-making | Which decisions does the system make without a human, and how are they reviewed? | GDPR Art. 22; ISO 42001, Annex A.5 (impact assessment); EU AI Act Art. 26 for deployers of high-risk AI from December 2, 2027 |
This overlap is the real lever for European companies. If you run an AI management system under ISO 42001, you have already documented your AI policy, roles, impact assessments, and data measures. If you have added the ISO 27090 threats to your risk register, you can evidence prompt manipulation and data poisoning with risk assessments you already have.
Our overview of the AI security standard ISO 27090 maps its nine attack types to Annex A. ISO 27090 is guidance without a certification; it supplies threats and countermeasures, not an audit of its own.
The EU AI Act regulates two of the seven topics by law: transparency of AI-generated content under Art. 50 and, from December 2, 2027, the obligations of deployers of high-risk AI under Art. 26. You have to comply with these obligations regardless of SOC 2. Our comparison of ISO 42001 and the EU AI Act explains where the law ends, where the standard begins, and when certification pays off.
What companies with ISO 27001 should do now
You do not need to rebuild anything yet. Three steps pay off whatever the final text says: describe controls by their outcome, keep the mapping to criterion numbers as a separate layer, and inventory your AI components.
The first matters most. Describe each control so that it makes sense without a number: "Each team lead reviews access rights to production systems quarterly; deviations are corrected within five business days." Keep the mapping to ISO 27001 Annex A and to the common criteria separate. When the numbers change, you change only that mapping, not the control and not the evidence.
Our article on ISO 27001 controls shows how to select the 93 Annex A controls and justify the selection in the Statement of Applicability. What else to do depends on where you start.
| Your situation | What makes sense now |
|---|---|
| First SOC 2 report in 2027 | Plan against the 2017 criteria and do not wait for the new version. Describe controls by outcome so that the later remapping stays small. |
| Annual Type 2 renewal | Ask your service auditor which criteria apply to your next engagement and how they treat a period that runs past December 31, 2027. Fill in the mapping column once the final criteria are out. |
| AI feature in your product | Inventory AI components: model, provider, data processed, owners. Link existing risk assessments under ISO 42001 or ISO 27090 instead of writing new ones. |
| Coding agents in development | Treat agents with write access as privileged accounts (ISO 27001 A.8.2) and route their changes through your existing change management (A.8.32). In our assessment, this is where the AI points of focus meet the change management criteria. |
| Many cloud and AI vendors | Extend your supplier assessment under ISO 27001 A.5.19 to A.5.23 to model providers: use of your data for training, retention periods, and notice of model changes. The 2017 criteria cover vendor risk under CC9.2. |
If you are still weighing whether a first report is worth it, our article on the benefits of a SOC 2 report for growing tech companies makes the case from a sales perspective.
Example: a Type 2 renewal across the 2027/2028 year-end
A fictional SaaS company in Munich with 140 employees has held ISO 27001 certification since 2024 and gives US customers a SOC 2 Type 2 report every year. Its review period runs from October 1 to September 30. The numbers are illustrative.
The next period, October 2027 to September 2028, runs past December 31, 2027. Whether the old or the new criteria apply depends on when the engagement is considered to begin, not on when the period ends. The company clarifies this with its service auditor in fall 2027, before it signs the engagement letter.
Its control matrix holds about 120 controls, each described by outcome and mapped separately to Annex A and the common criteria. If the new criteria apply, the team changes only the mapping column. The evidence from daily operations, such as access reviews and change tickets, stays the same. The company also runs a support assistant on a hosted language model; its risk assessment comes from the ISMS and is reused for the AI points of focus.
How Kertos maps ISO 27001 and SOC 2 to one control set
With Kertos, you maintain each control once and map it to every framework you need evidence for, such as ISO 27001, SOC 2, ISO 42001, and GDPR. If you are already certified to ISO 27001, SOC 2 builds on your existing controls and evidence instead of starting a second program. The mapping to Annex A and to the common criteria is its own layer on top of the same control and the same evidence. If the AICPA changes the numbering, that layer changes, not your control descriptions.
The platform collects evidence automatically through more than 100 integrations, so a Type 2 report rests on a continuous record instead of snapshots. You inventory AI systems on the same platform, and their risks, including the threats from ISO 27090, sit in the same risk register as your other information security risks.
An independent CPA firm issues the SOC 2 report itself. Kertos's certified experts prepare you for that examination and help you decide which AI components are in scope and how to evidence them. To see how your existing ISO 27001 ISMS covers SOC 2, book a demo.
Frequently asked questions
Is there a SOC 2 certification?
No. A SOC 2 examination ends with a report that a licensed, independent CPA firm issues under the AICPA attestation standards: Type 1 as of a point in time or Type 2 over a period. There is no certificate with a fixed validity period like ISO 27001, which is why customers usually ask for your most recent report.
When do the new SOC 2 criteria take effect?
If the AICPA approves the draft, the 2027 SOC 2 Trust Services Criteria apply to engagements beginning on or after December 31, 2027, with early adoption permitted. The AICPA accepts comments on the draft until January 15, 2027. It has not announced a date for the final criteria.
Do I need to change my SOC 2 program now?
No. Engagements that begin in 2027 will generally still use the 2017 criteria, and the draft can change after the comment period. What makes sense now is to describe controls by outcome, keep the mapping to criterion numbers as a separate layer, and inventory your AI components.
What is the difference between SOC 2 and ISO 27001?
ISO 27001 is an international standard for an information security management system and leads to a certification from an accredited certification body, valid for three years with annual surveillance audits. SOC 2 is a US attestation standard from the AICPA and leads to a CPA firm's report on the controls for the Trust Services Criteria categories you choose. Many controls overlap, so one shared control set can cover both.
What AI requirements does SOC 2 include?
The 2027 draft adds points of focus on AI governance, model monitoring, prompt manipulation, data poisoning, transparency of AI-generated outputs, training data management, and the risks of automated decision-making. Points of focus do not prescribe a specific control, but they steer what auditors ask. If you already handle AI risk under ISO 42001, ISO 27090, or the EU AI Act, you can use those measures as evidence.
