AICPA Draft Would Update SOC 2 Criteria for AI and Automated Systems

The AICPA’s proposed revisions would bring SOC 2 evaluations closer to the risks created by artificial intelligence and automation, while also reorganizing the criteria CPAs use to assess organizations’ information systems. The draft is open for comment through Jan. 15, 2027; if approved next year, it would apply to engagements beginning on or after Dec. 31, 2027, with early adoption permitted.

The proposal matters to organizations that rely on SOC 2 reports to explain their controls to customers and other report users, and to the CPAs who examine those controls. It could change the questions considered in an examination and how clearly a report communicates what a criterion is intended to address. The draft does not itself establish that any particular organization’s controls are adequate; that assessment remains tied to the organization and the engagement.

The American Institute of CPAs (AICPA) is seeking feedback on the proposed SOC 2 Trust Services Criteria update. SOC 2 is an examination conducted by a licensed, independent CPA under the AICPA Attestation Standards. It evaluates controls relevant to one or more of five areas: security, availability, processing integrity, confidentiality and privacy.

Reworking the criteria

The exposure draft proposes a targeted restructuring. It would consolidate concepts that overlap, remove criteria where appropriate, add criteria where the AICPA sees a need for more clarity or emphasis, and revise existing language. The stated aim is a more concise and consistent explanation of what each criterion is meant to cover.

That editorial work can affect how organizations and auditors interpret the framework in practice. When requirements overlap or their intent is hard to distinguish, teams may have difficulty mapping controls to criteria and explaining how those controls work. Clearer wording could make that process easier to follow across an examination and its resulting report. The proposal’s practical value will depend, however, on how the revised language handles real differences among systems, services and risks.

The AICPA says the changes are intended to keep the criteria relevant as organizations alter how they manage and process information. Amy Pawlicki, the institute’s vice president of assurance and advisory innovation, said the way organizations handle information and identify risk has changed substantially over the past decade. She said the proposed revisions are intended to improve report usability and help the criteria adapt to rapidly changing technology.

AI risks enter the focus

A central element of the draft is expanded attention to emerging technology, including AI. New points of focus address AI governance, model monitoring, prompt manipulation, data poisoning, transparency around AI-generated outputs, training data management and risks associated with automated decision-making.

These topics span the design and use of AI systems. Governance concerns how an organization oversees those systems; monitoring concerns how it identifies changes or problems in a model’s behavior. Prompt manipulation and data poisoning point to ways inputs can affect system outputs, while training data management and transparency raise questions about the information used and the results produced. Automated decision-making adds another area for scrutiny when systems make or shape decisions without the same kind of direct human review.

For organizations using AI or automation in services covered by a SOC 2 examination, these proposed focus areas may make it harder to treat technology oversight as a general statement of intent. Teams may need to explain how relevant risks are identified and how controls address them. Auditors, in turn, would consider whether those controls respond to the risks in scope. The draft describes areas for consideration; it does not prescribe a particular technology or say that every organization faces identical exposure.

Timing and implementation questions

The AICPA’s SOC 2 Working Group is accepting comments on the exposure draft through Jan. 15, 2027. The criteria remain proposed until approved. If approval comes next year, the planned effective date would cover engagements beginning on or after Dec. 31, 2027, and organizations could adopt the changes earlier.

That schedule gives affected organizations and CPA firms a defined window to assess the draft and identify language that may be unclear or difficult to apply. Feedback could help determine whether the revised criteria offer enough direction on new technology risks while remaining usable across different kinds of systems. The transition also gives teams a reason to review how they document controls and communicate them to report users, especially where AI, automated processing or model outputs are part of a service.

The next milestone is the close of the comment period, followed by the AICPA’s decision on whether to approve the proposed criteria. Until then, the draft signals the direction of the framework: clearer, more streamlined criteria, with explicit attention to technology risks that have become more prominent in how organizations process information and make decisions.