Third-party AI agents bypass identity controls in most enterprise environments

The scale of invisible AI agents

Findings from the 2026 State of Agent Security Report reveal a significant governance challenge for regulated institutions. In the environments studied, approximately 1,280 third-party products now embed AI agents. Of these, only 282 authenticate through single sign-on systems, leaving roughly 1,000 AI agents operating outside the visibility of identity infrastructure.

This gap does not result from deliberate concealment. Rather, it reflects a fundamental limitation: identity stacks can only govern what authenticates through them, and most embedded AI agents never do. The agents operate within third-party products that may already have access to enterprise systems, but the AI components themselves remain unregistered with central identity controls.

Implications for Saudi institutions

For organizations subject to NCA and SAMA frameworks, this presents a control gap. Third-party AI agents that bypass identity infrastructure cannot be inventoried, monitored for privilege escalation, or audited through conventional identity and access management tools. The agents may process sensitive data, make decisions, or interact with other systems without appearing in access logs or governance dashboards.

The challenge is architectural. Security controls designed for AI systems an organization explicitly deploys do not extend to agents embedded in software already in use. As AI capabilities become standard features in enterprise software, the number of ungoverned agents will grow unless institutions adapt their identity and access frameworks to detect and manage agents that do not authenticate in traditional ways.