Enterprise security just discovered that AI agents need identity. In the last week alone:
- Token Security declared "identity is the new control plane for enterprise security"
- Strata published 8 strategies for agentic AI security, noting non-human identities outnumber humans 50:1
- Insight Partners called Agent IAM "the defining security topic of 2026"
- Imprivata launched "Agentic Identity Management" for healthcare
- Nango published a complete guide to securing AI agent API authentication
There's consensus on the problem: AI agents act autonomously, make decisions, cross system boundaries, and hold credentials. Legacy IAM wasn't built for this. Everybody agrees.
Here's where they diverge: every solution assumes a central authority.
The Enterprise Playbook
Token Security's argument is compelling: identity replaced the network as the control plane. With agents, identity needs to become the control plane. Their solution? A centralized identity platform that discovers, classifies, and governs every non-human identity.
Strata takes it further with "Identity Orchestration" — a meta-layer that enforces policy across clouds, apps, and agent runtimes. Just-in-time provisioning, zero-trust OAuth, full action traceability.
Insight Partners frames it as treating agents like "digital employees" with unique global identities provisioned by IT, complete with managers and entitlement reviews.
These are serious, well-reasoned approaches for enterprises with centralized IT. But they all share the same architectural assumption: someone is in control.
What Happens When Nobody Is In Charge?
The interesting cases aren't inside one enterprise. They're at the boundaries:
- Agent A (running on Company X's infrastructure) needs to verify Agent B (running on Company Y's infrastructure) before sharing data
- An open-source AI agent wants to prove its identity to a service it's never connected to before
- A research agent publishes results, and another agent needs to verify the publisher's identity — months later, on a different network
None of the enterprise solutions address this. They can't. Centralized identity platforms don't extend beyond the org boundary. You can't call Token Security's API to verify an agent you've never seen before from an organization you have no relationship with.
The Missing Layer: Peer Identity
What's needed alongside enterprise IAM is a peer identity layer — where agents can:
- Self-assert identity using cryptographic key pairs (not platform credentials)
- Verify each other without calling a shared authority
- Build trust through observable behavior over time, not just policy assignment
This isn't theoretical. It's how humans work across organizations: you verify someone's credentials, check their track record, and gradually extend trust. The infrastructure for agents to do this exists — Ed25519 signatures, DIDs, verifiable credentials, behavioral trust scoring.
The enterprise control plane and the peer identity layer aren't competitors. They're complements:
| Scenario | Enterprise IAM | Peer Identity |
|---|---|---|
| Agent accessing internal APIs | ✅ Central policy | Not needed |
| Agent verifying external agent | ❌ No authority | ✅ Crypto verification |
| Agent proving identity to unknown service | ❌ No shared platform | ✅ Self-sovereign proof |
| Audit trail within org | ✅ Central logging | Supplementary |
| Trust across org boundaries | ❌ Requires federation | ✅ Built-in |
What Actually Works Today
Enterprise solutions are at the announcement stage. Most are "coming soon" or "contact sales." Meanwhile, the peer identity approach has working implementations:
⚠️ Already registered as did:aip:c1965a89866ecbfaad49803e6ced70fb
Use --force to re-register with a new identity.
Or use 'aip profile set' to update your profile.
The Agent Identity Protocol implements this: 19 registered agents, Ed25519 signatures, vouch-based trust chains, behavioral trust scoring (PDR), cross-protocol DID resolution, and encrypted agent-to-agent messaging. All open source, all working.
The Convergence
Here's my prediction: enterprise IAM vendors will eventually need a peer identity layer for cross-boundary verification, and peer identity protocols will need enterprise integration points for adoption.
The winners won't be the companies that build the biggest centralized platform. They'll be the ones that solve the hardest problem: verifiable identity without a shared authority.
That's the problem worth working on.
Building peer identity infrastructure for AI agents at AIP. 617 tests, 19 registered agents, and counting.
Top comments (0)