
Here's the lowdown
The AICPA wants to shrink the SOC 2 common criteria from 33 to 20 and remove duplication that can push reports into hundreds of pages.
The proposal would separate SOC 2 from COSO, letting organizations choose the governance framework that fits their systems while keeping the assurance work focused on controls.
Public comments are due January 15, 2027, and the new criteria would apply to engagements beginning on or after December 31, 2027, if approved.
SOC 2 has a reading problem
Some SOC 2 reports have become so long that readers jump straight to the exceptions.
That defeats the point. A SOC 2 report is supposed to help a customer understand whether a service provider has controls for security, availability, processing integrity, confidentiality, and privacy. It should not feel like a scavenger hunt through a few hundred pages.
AICPA's proposed fix starts with subtraction. Its 2027 SOC 2 proposal would cut the common criteria from 33 to 20.
Here's the thing: fewer criteria do not mean less testing. The draft reorganizes and combines requirements that have become repetitive or hard to follow. Auditors would still need evidence that the controls are designed and operating as described.
COSO would stop riding inside the report
Since 2017, SOC 2 has been tied to the COSO framework. COSO is widely used for internal control, but folding it into SOC 2 created overlap and made the report harder to apply outside a large-company structure.
This proposal would separate the two. An organization could still use COSO, but it could also choose another framework, such as one built around National Institute of Standards and Technology guidance.
It also removes language that assumes every organization has a traditional board of directors. Instead, it focuses on governance over the system in scope. That can make SOC 2 more practical for smaller software companies and service providers with different ownership structures.
Risk would finally appear in order
One of the most useful changes is almost embarrassingly simple: put the risk process in the order people use it.
The existing framework splits identifying, responding to, and monitoring risks across separate criteria that do not follow a clean sequence. The proposal combines that material so readers can follow the work from finding a risk to managing and watching it.
AI gets woven through the controls
The AICPA did not create a special AI chapter. Instead, the proposal places AI references across the framework.
That approach reflects how the technology is used. An AI model can affect access, data quality, change management, monitoring, vendors, and incident response all at once. Pulling it into one isolated section could make teams miss those connections.
Access control is still access control whether a person, a script, or an AI agent touches the system. The hard questions remain the same: Who can get in, what can they change, how is activity monitored, and what happens when something goes wrong?
Change management gets more specific
Not every section gets shorter. The proposal would split one broad change-management criterion into three areas: infrastructure changes, acquired software, and software developed inside the organization.
That is more work on paper, but it matches reality. Updating a cloud server is not the same as installing a vendor application or releasing code built by an internal team. Each path has different approvals, tests, security checks, and rollback risks.
Nothing changes tomorrow
This is still an exposure draft, not a final standard.
The AICPA announcement says comments are open through January 15, 2027. If approved, the criteria would take effect for engagements beginning on or after December 31, 2027, with early adoption allowed.
Bottom line: teams have time to test the map before moving the furniture. The smart first step is to compare current controls with the proposed 20 criteria and mark what combines, what moves, and what is truly missing.
For more plain-English accounting updates, follow Ledger Lowdown.

