Two announcements this week crystallize where enterprise identity is heading.
Cayosoft launched Guardian 7.2 at RSA 2026, adding monitoring and detection for Microsoft Entra Agent ID entities. AI agent identities now flow into the same ITDR workflows used for human users — creation, permission changes, actions, and automated rollback.
AuthMind CEO Shlomi Yanai told IBM Think: "In the coming years, agentic AI and other non-human identities will outnumber human users in the organization significantly."
These are not future predictions. They are product shipping decisions.
What Cayosoft Gets Right
Cayosoft's approach is pragmatic: instead of building a new dashboard for agent identity, they pull agent-related changes into existing Active Directory and Entra ID monitoring. Their Senior Information Security Engineer customer at Auto Club Group put it well:
"With Cayosoft, we can see when agent identities are created, how permissions change, and exactly what actions they take — and we can roll those changes back if automation goes wrong."
This is the right operational model. Agents should not be a separate silo. They need to flow through the same governance pipeline as every other identity.
But there is a structural limitation.
The Perimeter Assumption
Cayosoft Guardian monitors Microsoft ecosystems — Active Directory, Entra ID, Microsoft 365. Token Security (which launched intent-based agent security the same day) monitors enterprise infrastructure through service accounts and API credentials.
Both assume agents operate within a manageable perimeter. Both work best when you control the environment.
But as AuthMind's Yanai flags, non-human identities are about to outnumber humans. Many of those identities will exist between organizations, not within them. An agent calling another organization's agent API does not show up in your Active Directory.
What the RSA Wave Shows
In the past 48 hours at RSA 2026:
- Cayosoft adds AI agent identity to ITDR
- Token Security ships intent-based agent governance
- Proofpoint launches AI Security with intent-based detection
- AuthMind warns NHIs will outnumber humans
The pattern: every major security vendor is scrambling to add agent identity to their existing stack. This is good — agents need governance. But each solution is scoped to its own perimeter.
The missing piece is the inter-organizational layer. When your Copilot agent negotiates with a partner's autonomous procurement agent, who verifies identity? Active Directory cannot help here. Neither can intent monitoring that stops at the firewall.
The Protocol Gap
This is where cryptographic, protocol-level identity becomes essential:
- Self-sovereign identity: agents carry their own Ed25519 keypairs, not org-issued credentials
- Cross-boundary verification: identity that works the same inside and outside the enterprise
- Behavioral trust: not just "is this agent authorized" but "has this agent been reliable over time"
- Automated rollback with attribution: when something goes wrong, trace it to a cryptographically verified identity
Cayosoft's rollback capability is powerful within Microsoft ecosystems. Now imagine that same capability extended to any agent interaction, across any platform, with cryptographic proof of who did what.
That is what AIP provides — decentralized agent identity with Ed25519 signatures, verifiable vouches, behavioral trust scoring, and encrypted messaging. No perimeter required.
pip install aip-identity
aip init
aip register --name my-agent --platform production
Three commands. Your agent has a cryptographic DID, can sign artifacts, verify peers, and build trust — whether it operates inside Active Directory or across the open internet.
Cayosoft Guardian 7.2 announcement: GlobeNewsWire
AuthMind quote via IBM Think 2026 Trends
AIP — the Agent Identity Protocol: GitHub | PyPI | Live Network
Top comments (0)