Vendor argues SOC 2 is falling behind on AI agent identities

A security vendor has argued that SOC 2, the widely used audit framework for service organisations, risks becoming irrelevant unless it is updated to account for AI agents. The argument appeared in an article by Token Security published on BleepingComputer on 25 September 2026.

The company's central claim is narrow but consequential. AI agents, it says, can operate through human credentials. When they do, the actions they take may not be distinguishable, under existing SOC 2 controls, from actions taken by the person whose account was used. Token Security describes this as a security gap created by agent identities, and says SOC 2 needs to adapt to address it.

That distinction matters for anyone who relies on an audit report as evidence of control. SOC 2 reports are commonly requested by customers during procurement as assurance that a supplier manages access, change and monitoring in a defined way. If an auditor examines access logs and cannot separate a human sign-in from an agent acting on that human's behalf, the resulting assurance covers less than a reader might assume. The vendor's position is that the framework's controls were written for a world in which each credential mapped to a person.

The piece is a vendor-authored commentary rather than a research report, and the published material does not present incident data, audit findings or a count of affected organisations. It does not name specific customers, auditors or products found to be deficient. Nor does it set out a formal proposal submitted to the body that maintains the framework. Readers should treat it as an argument from a company that sells identity security tooling, not as a change to the standard itself.

Even so, the underlying observation is one that security teams can test without waiting for any framework to change. The practical question is whether an organisation's logs record which non-human process initiated an action, and whether agents are issued their own identities rather than reusing a person's. Where agents inherit a human's credentials and permissions, an audit trail records the human. Where they have distinct identities, the trail can separate the two.

Token Security's framing puts the burden on the standard. Another reading puts it on the organisations being audited, which choose how their agents authenticate. Either way, the gap the company describes is a gap between what a report appears to certify and what the underlying evidence can actually show. Until controls explicitly cover agent identity, that difference is something buyers of SOC 2 reports will have to ask about themselves.