Revelations AI

Revelations AI blog

SOC 2 change management for AI agents: proving who approved

Revelations Consulting · 2026-11-19

One question sits at the center of SOC 2 change management for AI agents: when an agent makes a change, can you show, from records captured at the time, who independently approved it? The AICPA describes SOC 2® as “an examination performed by a licensed, independent CPA under the AICPA Attestation Standards that evaluates controls relevant to the security, availability, processing integrity, confidentiality or privacy of an organization’s systems.” The criteria in force are the final 2017 Trust Services Criteria (With Revised Points of Focus – 2022), control criteria established by the AICPA’s Assurance Services Executive Committee (ASEC). The AICPA has issued exposure drafts of proposed revisions to the Trust Services Criteria and Description Criteria; these proposed 2027 criteria are not in effect.

Can auditors even ask about AI agents right now?

The 2017 Trust Services Criteria are control criteria for evaluating and reporting on controls over “information and systems used to provide products or services.” The AICPA sources cited here do not say what auditors will specifically ask about AI agents.

The Gated Agent Deployment page names the risk: “Most ‘multi-agent review’ is one process wearing several hats. When an auditor or a post-incident review asks who independently approved a change, there’s no answer you can prove.” That is the page’s framing, not a quote of the criteria.

What does “independent approval” actually mean in practice?

The AICPA sources cited here do not define independent approval, and the full PDF of the 2017 Trust Services Criteria requires an AICPA login. That leaves a check you can run, not a definition to quote.

The Gated Agent Deployment page describes one design: approvals need “an author, a reviewer on a different model, and a person’s passkey for production.”

To test your own setup, take one gated change and list the identity that wrote it, the identity that reviewed it and the identity that approved production. If two of those turn out to be the same process, service account or signing key, your records show one process wearing several hats, the risk the page names.

Are the proposed SOC 2 criteria already in effect?

No. The AICPA’s Assurance Services Executive Committee (ASEC) has issued exposure drafts of proposed revisions to the Trust Services Criteria and Description Criteria. The AICPA says its SOC 2 Working Group is seeking public comment through Jan. 15, 2027. If approved next year, the proposed 2027 criteria would be effective for engagements beginning on or after Dec. 31, 2027, with early adoption permitted.

According to the same AICPA release, the proposed new points of focus “address issues such as AI governance, model monitoring, prompt manipulation, data poisoning, transparency of AI-generated outputs, training data management, and the risks associated with automated decision-making.” The proposal also consolidates duplicative concepts, removes criteria where appropriate, adds new criteria where greater clarity or emphasis is needed, and revises existing criteria language.

What evidence would show who approved an AI agent’s change?

A reasonable expectation, not a quote of what auditors test, is that records show who wrote the change, who reviewed it and who approved the production step, captured when the change happened. Rebuilding approval records after the fact is not the same as evidence captured at the time.

A check you can run this week: pick one recent change made by an agent and try to show who independently approved it, using only records written at the time. Check whether the reviewer ran on a different model or identity than the author, and whether a person approved the production step. If you cannot answer from those records, that is the gap the Gated Agent Deployment page names: no answer you can prove.

The page says its handover includes “an evidence pack mapped to the controls your auditor tests.” Its offer also includes a “Reviewer red-team with planted mistakes”, and its “Test the reviewers” stage says: “Plant mistakes and check the reviewers catch them, so approvals mean something.”

The apps built on it are in early access. Which agent change could you not yet prove an independent approval for? Tell us on a call: book a 30-minute call.

Frequently asked questions

What is SOC 2?

The AICPA describes SOC 2® as “an examination performed by a licensed, independent CPA under the AICPA Attestation Standards that evaluates controls relevant to the security, availability, processing integrity, confidentiality or privacy of an organization’s systems.” The criteria in force are the final 2017 Trust Services Criteria (With Revised Points of Focus – 2022), established by the AICPA’s Assurance Services Executive Committee (ASEC).

Are the proposed SOC 2 criteria in effect yet?

No. The AICPA describes the 2027 SOC 2 Trust Services Criteria as proposed, with public comment through Jan. 15, 2027. If approved next year, the AICPA says they would be effective for engagements beginning on or after Dec. 31, 2027, with early adoption permitted.

Do auditors already ask about AI agents?

The AICPA’s proposed 2027 criteria would add AI points of focus, including AI governance, model monitoring and prompt manipulation, but they are proposed, not in effect. The AICPA sources cited here do not say what auditors will specifically ask about AI agents.

Does the reviewer have to be a different model?

The Gated Agent Deployment page says the reviewer is “deliberately on a different model from the author.” The AICPA sources cited here do not say whether a reviewer must run on a different model, and the full PDF of the 2017 Trust Services Criteria requires an AICPA login.

Can AI agents that run Terraform fit a SOC 2 change process?

The Gated Agent Deployment page says: “We’re model-neutral. The reviewer is deliberately on a different model from the author, and we plug in the tools you already use.” Its typical first agents include “Terraform plan and apply” and “infrastructure pull requests.”

This isn't compliance advice. Primary source: AICPA Proposes Update of SOC 2® Trust Services Criteria.