SOC 2 compliance depends on several assumptions about user identities and access controls that no longer hold true with AI agents. These agents often operate under borrowed human credentials, lack explicit approval or ownership, and their actions are logged as if performed by humans, obscuring their distinct security risks. This undermines the effectiveness of controls related to account approval, access reviews, offboarding, vendor management, and segregation of duties. For example, agents can continue operating after a human leaves, and vendor reviews cannot account for agents without clear organizational ownership. The current SOC 2 Trust Services Criteria do not explicitly require organizations or auditors to treat AI agents as separate identity classes, allowing these gaps to persist. Token Security highlights the need for SOC 2 to evolve to detect, identify, and remediate AI agent access to maintain relevance and security assurance.

The presence of AI agents that use human credentials and lack explicit identity management weakens SOC 2 controls, potentially allowing unauthorized or unmonitored actions to occur without detection during audits. This can lead to security blind spots where AI agents perform activities that are indistinguishable from legitimate human actions, undermining trust in SOC 2 compliance reports. Controls for offboarding, vendor review, and segregation of duties are particularly affected, increasing the risk of persistent unauthorized access and insufficient oversight of automated processes. Organizations relying on SOC 2 for customer trust may face increased risk exposure if these gaps are not addressed.

Currently, SOC 2 does not explicitly require treating AI agents as distinct identities, and existing controls may not detect or manage agent activities effectively. Organizations should evaluate their environments for AI agent presence and consider implementing tools that can identify and assign identities to AI agents, such as those offered by Token Security. Remediation involves restricting agent access to only the tasks they are intended to perform and ensuring offboarding processes include agent deactivation. Until SOC 2 criteria evolve, organizations must supplement compliance efforts with additional controls and monitoring tailored to AI agents. Patch status is not applicable as this is a compliance framework issue rather than a software vulnerability.