Short answer
The AICPA has published an exposure draft proposing a revision to the Trust Services Criteria — adding AI-specific points of focus and tightening language on third-party service providers and continuous monitoring. The five criteria (Security, Availability, Processing Integrity, Confidentiality, Privacy) are unchanged. The public comment period is open now and closes before the AICPA finalises the text for a 2027 effective date. Verify the exact comment deadline on the authoritative AICPA ASEC exposure-draft page before relying on it in a report or contract.
Through 2026 the AICPA worked around a problem the Trust Services Criteria didn't anticipate: AI running inside the service. September's Technical Question and Answer Section 9561 was the stopgap — non-authoritative guidance telling auditors to apply the existing criteria to AI. The 2027 revision is the authoritative follow-through, taking that practice and codifying it as revised points of focus. This explainer covers what the exposure draft actually changes, the comment deadline, and what an AI SaaS team should do between now and the effective date.
If you're earlier in the SOC 2 journey, start with the SOC 2 framework page and the Trust Services Criteria glossary entry.
What's in the exposure draft
The draft keeps the structure of the 2017 TSC: five criteria, Common Criteria (CC1–CC9), category criteria for the optional four, and points of focus underneath each criterion. What changes is the points of focus — the illustrative sub-bullets that explain how a service organisation might meet each criterion.
| Area | Current (2017, rev. 2022) | Proposed 2027 |
|---|---|---|
| Five criteria | Security, Availability, Processing Integrity, Confidentiality, Privacy | Unchanged |
| AI governance | No specific points of focus — interpreted via TQA 9561 | New points of focus under CC3 (risk assessment) and CC5 (control activities) covering AI model inventory, intended use, and oversight |
| AI data governance | Covered implicitly under C1 and P-series | New points of focus explicitly addressing training-data provenance, data minimisation in prompts, and no-training opt-outs |
| Model + prompt change control | Covered by generic CC8.1 change management | New point of focus naming system prompts, prompt templates, and model-version pinning |
| Third-party model providers | Under CC9.2 vendor management, no AI-specific language | Sharpened language on subservice organisations, including carve-out vs inclusive treatment for AI providers |
| Monitoring and incident response | CC7.2/7.3 | New points of focus explicitly naming prompt injection, output harm, and model-provider outage scenarios |
| Continuous monitoring + risk reassessment | CC3.4, CC4 | Tightened language on cadence and triggers for risk reassessment when AI capabilities change |
Important nuance for implementers: points of focus are not mandatory controls. The criteria themselves are the requirements; points of focus illustrate how to meet them. Adding AI points of focus doesn't automatically force every service organisation to implement AI controls — only those whose AI usage is in SOC 2 scope.
Timeline and comment deadline
- Exposure draft published: late 2026 by the AICPA Assurance Services Executive Committee (ASEC).
- Public comment period: open now. The exact closing date is published on the AICPA ASEC exposure-draft page — verify directly, as the window typically runs 60–90 days and may be extended. Treat any calendar date here as subject to AICPA confirmation.
- Final text: expected in 2027 after ASEC reviews comments and incorporates revisions.
- Effective date: expected in 2027 for reports covering fiscal periods ending on or after that date. The AICPA historically allows a transition window where either the current or revised criteria may be applied.
Reports issued today under the 2017 criteria (as revised in 2022) remain acceptable until the stated effective date. Enterprise buyers will not reject a current-criteria report during the transition window.
How to comment on the exposure draft
- Download the exposure draft from the AICPA ASEC page.
- Review the proposed points of focus most relevant to your service — CC3, CC5, CC7, CC8, CC9, C1, P4 for most AI SaaS.
- Draft specific line-level feedback: cite the point-of-focus number, quote the current wording, propose the edit, and give a 2–3 sentence rationale.
- Submit via the AICPA comment channel named on the exposure-draft page before the deadline.
The most valuable comments come from practitioners with concrete implementation experience. If a draft point of focus would be impractical to evidence (for example, "maintain a comprehensive catalogue of all AI model inputs" without defining comprehensiveness), say so and propose a narrower, auditable alternative. The AICPA weighs implementability heavily.
What AI SaaS teams should do between now and the effective date
- Build to TQA 9561 today. Our TQA 9561 breakdown lists the controls auditors are already asking for under the current criteria. The 2027 exposure draft is a strict superset of that guidance — nothing changes direction.
- Stand up an AI/model inventory. One row per model touched by the service, with provider, data flows, model version, and intended use. Required under the proposed CC3 points of focus; also required by ISO 42001.
- Change-control prompts, tools and model versions. System prompts, prompt templates and model versions in source control with reviewed PRs — see SOC 2 controls for AI agents and MCP.
- Formalise your vendor review for the LLM provider. Current SOC 2 or ISO 27001 report on file, DPA with a no-training clause, annual re-review.
- Document incident response for AI-specific scenarios. Prompt injection, output harm, model-provider outage, data leakage via tool chaining.
- Start the Type 2 observation window now against current criteria. By the time the final 2027 text lands, your observation period will have run under the current criteria with controls that already meet the proposed revisions. No rework.
What is NOT changing
- The five criteria — Security, Availability, Processing Integrity, Confidentiality, Privacy — stay as-is.
- SOC 2 remains an attestation, not a certification. Only licensed CPA firms can issue reports (see how to verify your SOC 2 auditor).
- Type 1 vs Type 2 structure is unchanged.
- The two optional reporting formats (SOC 2 and SOC 3) remain.
- No new mandatory criterion for AI — AI is covered via points of focus under existing criteria.
Frequently asked questions
What is the proposed 2027 SOC 2 Trust Services Criteria revision?
The AICPA ASEC (Assurance Services Executive Committee) published an exposure draft proposing updates to the 2017 Trust Services Criteria. The headline changes are new AI-specific points of focus under existing CC-series criteria and sharpened language around third-party service providers and continuous monitoring. No new criterion is being added — the five remain Security, Availability, Processing Integrity, Confidentiality and Privacy. The comment period is open now and closes before the AICPA finalises the revised criteria for an expected 2027 effective date. Check the AICPA exposure-draft page for the exact deadline, which should be verified against the authoritative source before any public reliance.
Does this replace the 2017 TSC?
No — it is a revision to the existing TSC, not a replacement. The structure (five criteria, Common Criteria + category criteria, points of focus) stays. The revised document amends points of focus and clarifying language. Reports issued today under the 2017 criteria remain valid until the new revision's effective date lands.
How does the exposure draft relate to AICPA TQA 9561?
TQA Section 9561, released September 2026, was the non-authoritative "how do we handle AI under the existing TSC?" guidance. The 2027 revision is the authoritative follow-through — the ASEC taking what practitioners have been doing informally and codifying it as points of focus. If you built your AI controls to meet TQA 9561 (see our breakdown of SOC 2 AI AICPA TQA 9561), you will be substantially aligned with the exposure draft.
Should I wait to start SOC 2 until the final 2027 criteria are published?
No. Enterprise buyers still ask for SOC 2 today, and reports issued under the current (2017, revised 2022) criteria remain acceptable until the AICPA's stated effective date for the revision. The practical move is to start SOC 2 Type 1 now against the current criteria and layer in the proposed AI points of focus as controls so the Type 2 cycle lands on the revised criteria without rework.
How do I submit a comment on the exposure draft?
The AICPA accepts comments through its standard exposure-draft process on the ASEC page. Comments are read publicly and influence the final text. For AI startups the most valuable comments are concrete: show the auditor-side ambiguity in the current draft points of focus and propose plain-language edits. Keep comments evidence-oriented — regulators and practitioners weigh practical implementability heavily.
When does the revised TSC take effect?
The expected effective date is in 2027, with an application period for reports covering fiscal periods ending on or after the effective date. The AICPA typically allows a transition window where either the old or revised criteria may be applied. The exact dates are being finalised through the exposure draft process — verify the authoritative source before relying on specific calendar dates.
Get SOC 2 ready for 2027 without rework
LowerPlane maps SOC 2 and ISO 42001 to one AI evidence base, so controls built for today's criteria carry forward unchanged. See SOC 2 for AI startups.