Everyone wants agents. Nobody wants to inventory them. Vendors have noticed, and the new enterprise control plane is basically: ‘show me the delegation chain.’
AI agents don’t break security because they’re evil. They break security because they’re fast, ephemeral, and authenticated with credentials your organization can’t even find. The industry response is predictable: rename IAM, add ‘agentic,’ and sell you an audit trail.
Agents Are Delegated Identities, Not Cute Features
Enterprise security has a long tradition of discovering new categories of ‘identity’ only after they’ve already multiplied into an uncountable swarm. We did it with service accounts. We did it with API keys. Now we’re doing it with agents.
Orchid’s pitch is blunt: agents are delegated identities, so if you can’t see the chain of delegation — who authorized what, on whose behalf, under what conditions — you can’t govern anything. HashiCorp is making the same point from a different angle by adding an agent registry concept to Vault: track agent activity separately from humans and traditional non-human identities, especially in on-behalf-of flows.
Control Planes Are Converging
JumpCloud is describing “Agentic IAM” as a unified control plane for human, non-human, and agentic governance, including discovery of agents and even locally running resources like MCP servers. ServiceNow is framing the problem as identity and permission sprawl accelerating faster than enterprises can answer basic questions about who approved access and whether it’s still valid — and selling a single graph to map identities, permissions, and assets.
What the CSA Stats Should Make You Do (Today)
- Inventory: you can’t rotate what you can’t see.
- Short-lived credentials: treat long-lived agent tokens like open doors.
- Delegation-aware logging: “agent did X” is useless without “on behalf of Y.”
- Procurement pressure: buyers are going to ask for governance evidence the same way they ask for SOC2, ISO, and FedRAMP.
The Singularity Soup Take
The agent future is arriving as a paperwork future. You wanted autonomous coworkers. Congratulations: you now have autonomous audit requirements. Your participation in identity governance is becoming increasingly non-optional.
What to Watch
- Whether MCP (and other agent interop) gets treated as a security boundary with formal controls, not a convenience layer.
- Which vendor ends up owning the “agent registry” primitive across stacks (and how fast consolidation starts).
- Whether regulators start treating ‘delegation traceability’ as a minimum bar after the first big agent-enabled incident.
Sources
Cloud Security Alliance — "Agentic AI Governance: Addressing the Non-Human Identity Security Gap (May 2026)"
HashiCorp — "Announcing native AI agent support in HashiCorp Vault"
Orchid Security — "Orchid extends Identity Control Plane in response to agentic ‘dark matter’"
ServiceNow Newsroom — "ServiceNow launches Autonomous Security & Risk (Armis + Veza integration)"
JumpCloud — "JumpCloud launches Agentic IAM to govern the AI lifecycle"