AI agent governance

Inventory every AI agent, the identity it acts as and what it can reach, with checks mapped to the OWASP Top 10 for Agentic Applications.

Updated 29 October 2026

Online version, always current: https://cloudvision360.com/docs/cloudvision360/ai-agent-governance/

AI agents act with real identities and real permissions. CloudVision360 governs them like any other privileged identity.

Every agent, every platform

Inventory AI agents across Microsoft Entra, Copilot Studio, Azure AI Foundry, Amazon Bedrock and Vertex AI — the identity each agent acts as, and what that identity can reach.

The risks it checks for

Over-privileged agents — an agent whose identity can do far more than its task needs.
Ownerless agents — no accountable owner.
Orphaned agents — left behind after a project ended.

Checks are mapped to the OWASP Top 10 for Agentic Applications, so the findings line up with an emerging industry standard.

Governed like other identities

Agent risks appear as findings with owners, risk scores and the same triage and remediation paths as the rest of your identity inventory.

Use it

CloudVision360 AI agents page: agents, admin or organization-wide access, who can grant access, and agents without an owner — Security → AI agents: every AI agent and service in your clouds, the identity it uses, what it can reach and who owns it. (Sample data.)

Open Security → AI agents to review agents by platform, privilege and owner, and to act on over-privileged or orphaned agents. For governing the AI that CloudVision360 itself uses, see Ask CloudVision and the AI gateway.