AI agents reach every IAM job: what seven vendors announced in 2026

This post covers announcements and documentation dated from May to September 2026, using pages read on 27 and 29 September 2026.

What was announced, newest first?

• 22 September 2026, Okta: updates to Okta for AI Agents. Resource Access Certifications, which review agent permissions over time, are listed as available now; Agent Gateway and Shadow AI Agent Discovery for Endpoints are planned for Q3 2026.
• 21 September 2026, Lumos: MCP Governance, which lists the MCP servers and tool calls employees' agents use, records each call with the human identity and the policy decision, and enforces policy before actions run.
• 4 August 2026, SailPoint: SailPoint Identity Security, combining Agentic Fabric, now generally available, for AI and machine identities with Human Fabric.
• 28 July 2026, Saviynt: Zuma, an AI identity security platform with Zuma Insights, Zuma Access and Zuma Governance.
• 24 June 2026, Microsoft: a documentation page dated that day describes Entra Agent ID as generally available for all Microsoft Entra customers, with extended security features requiring Microsoft Agent 365, which comes in Microsoft 365 E7 or as an add-on.
• 15 June 2026, C1: the C1 Autonomous Worker, an agent that runs tasks such as revoking unused admin access and building access reviews with the operator's own permissions.
• 27 May 2026, Ping Identity: Programmable Identity (MCP, CLI and APIs), Agent Discovery and Governance, and Privileged Access for Desktop Agents, with availability that may vary by product, region or deployment model.

Silverfort also describes AI agent security, including discovery through read-only APIs across Entra ID, Okta, AWS, Azure and GCP and an MCP gateway; its page gives no launch date.

Are these the same kind of feature?

No. Two different ideas share the word "agent". Some features govern AI agents as identities: who they are, what they can reach, what they did. Others are agents that do IAM work: C1's Autonomous Worker runs governance tasks, and Lumos describes agents that help with access reviews. A buyer asking "does it handle agents?" should say which of the two they mean.

Which IAM job owns an agent?

• Its identity and sign-in: directory and single sign-on. Microsoft's Entra Agent ID gives agents identities with Conditional Access and ID Protection for agents.
• Its permissions over time: lifecycle and governance. Okta's Resource Access Certifications and SailPoint's Agentic Fabric sit here.
• Its tool calls at runtime: a new control point. Lumos MCP Governance and Silverfort's MCP gateway sit here, and Okta lists Agent Gateway as planned.
• Its credentials: privileged access, as for any machine identity. See the OWASP NHI Top 10 lesson.

What should you ask before relying on an agent feature?

• Is it generally available or planned? Okta lists two items as planned; Ping says availability may vary.
• Which licence does it need? Microsoft states that extended Agent ID features need Microsoft Agent 365.
• Does it find agents nobody registered? Okta's endpoint discovery is planned; Silverfort describes discovery through cloud and identity provider APIs.
• Is every agent tied to an accountable human? Lumos records the human identity with each call, and Silverfort describes binding each agent to a human identity.