AI agents are multiplying fast. See what 150 identity pros said about sprawl, shadow AI, and governance at Identiverse 2026.
At Identiverse 2026, Microsoft Security hosted a Power Breakfast that brought together 150 identity professionals across 10 simultaneous roundtable discussions. Participants came from industries including financial services, healthcare, government, and energy, and represented every stage of AI adoption.
We asked participants what they’ve built to secure agents, what they’ve rolled back, and where control is breaking down.
“The pace has been crazy fast. We have thousands of agents. Most of them are unmanaged.”
Roundtable participant, Identiverse 2026
What your peers are saying about securing agents
Agent sprawl is already here and bigger than most teams realize
Nine in 10 roundtables described unmanaged agent sprawl not as a future risk, but as a present reality. What started as dozens of agents became tens of thousands in months or even weeks. No single audit could capture the full picture. Practitioners described discovering the true number by accident, often finding far more than they expected: “I was doing a demo one day, I looked at our corporate tenant, and we had like 44,000 agents. I just couldn’t believe it, after so little time.”
Agents are proliferating across every cloud, SaaS platform, and vendor environment simultaneously
Nine in 10 roundtables reported agents running across SaaS, multicloud, and third-party environments with no governance visibility. Eight in 10 said shadow AI is already present in their organizations, running across platforms that no single governance layer currently covers and tracked only through network logs, if tracked at all. The challenge isn’t unique to any one platform: every SaaS vendor is now shipping with an agent, every cloud provider has a builder framework, and every team has a developer standing up something new.
“When everybody has a single pane of glass, nobody has a single pane of glass.”
Roundtable participant, Identiverse 2026
Assigning human ownership to agents becomes complex when people change roles or leave organizations
Nine in 10 roundtables raised the ownerless-agents problem. Agents get built, tied to their creator’s identity, and keep running long after that person has changed roles or left the organization—unreviewed, over-permissioned, and undetected. This results in agents with no clear accountability, stale permissions, and no obvious path for review, reassignment, or decommissioning.
Agent-to-agent interactions are where control fails hardest
Eight in 10 roundtables identified agent-to-agent chains as the most difficult security challenge they had encountered. Multiple teams rolled back agent-to-agent deployments after finding they could not maintain consistent governance across the chain.
“I can control point A to point B, but point B needs to talk to point C, and that’s where context was lost. I can control permissions. I can control authentication. I do not know how to control the impact.”
Roundtable participant, Identiverse 2026
How Microsoft Entra Agent ID can help
Microsoft Entra Agent ID is the identity foundation for every AI agent, and it’s where your control starts. Every governance, access, and lifecycle decision flows through agent identity. Without it, none of the controls below are possible. Microsoft Agent 365 builds on that foundation as the unified control plane, giving IT and security teams a single place to see and act on every agent across the organization.
If you have similar pain points around managing agents, here are three actionable starting points for securing them:
1. Build your inventory before you do anything else
You cannot govern what you cannot see. Start here.
The Microsoft Entra admin center shows every agent identity in your tenant across Copilot Studio, Microsoft Foundry, and third-party platforms in one view. Start in Microsoft Entra ID > Agents > Agents overview to get a full picture of the total number of agents with identities, how many were recently created, how many are active, and how many are unmanaged. Then go to Microsoft Entra ID > Agents > Agent identities and run your first audit. Agents without an Agent ID appear as classic agents with no governance controls attached. That list is your backlog and needs to be addressed right away to prevent sprawl.
For agents running outside the Microsoft ecosystem, register them using the Agent 365 CLI and SDK or federated identity credentials. You do not need to migrate them; you need to get them into the registry so they are visible and can be governed.
For third-party agents, running the Agent 365 SDK assigns each agent an agent identity blueprint, an agent identity, and a sponsor from the start, helping ensure that agents are trackable and more secure. A blueprint is a template that defines permissions, policies, and metadata for every agent instance created from it. It provides centralized control, allowing you to manage or disable agents created from it at scale. Skipping blueprints and creating agents as ad hoc service principals is how you end up with the sprawl the roundtable participants described.
- Get started with Microsoft Entra Agent ID
- Plan your agent identity architecture
- Agent identity blueprints
2. Assign an owner and a sponsor to every agent, then automate what happens when they leave
The orphaned-agent problem practitioners described has a direct solution. Every agent identity in Microsoft Entra Agent ID requires a sponsor (the person accountable for what the agent does) and an owner (the person responsible for its technical management). Assign both at creation time, at the blueprint layer. If you have existing agents with neither, start there.
When someone leaves your organization, Microsoft Entra lifecycle workflows can automatically trigger an ownership review for every agent associated with that person. You define the escalation path and the window: reassign, pause, or decommission. This replaces the manual scripts your team has been running for service accounts.
- Set up recurring access reviews for agent identities. Scope them to surface active permissions, usage signals, and business justification—not just whether the agent still exists. Agents that cannot be justified should be decommissioned, not left running.Best practices for Microsoft Entra Agent ID
- Govern agent identities and lifecycle
- Lifecycle Workflows overview
- Access reviews for agents overview
3. Scope permissions tightly, and apply Conditional Access and RBAC at the blueprint level
Start with enumerated scopes on every blueprint: only the specific delegated permissions the agent needs, nothing more. This is not optional. Agents do not self-restrict.
“An agent doesn’t self-restrict. It’s going to do whatever it feels like it needs to do.”
Roundtable participant, Identiverse 2026
For agents acting on behalf of a user, use the on-behalf-of flow so that user-level access policies apply. For autonomous agents, use the client credentials flow, scoped narrowly. Do not grant application-level permissions when delegated permissions accomplish the same task.
Apply Conditional Access policies at the blueprint level, not agent by agent. Every agent instance created from that blueprint inherits the policy automatically. This is how you extend Zero Trust controls from your workforce to your agents. Next, Microsoft Entra ID Protection analyzes agent behavior to detect anomalies such as unusual access patterns, spikes in activity, or interactions with unfamiliar resources. When an agent is flagged as risky, those signals automatically trigger Conditional Access policies to block or restrict access in real time, moving you from static policies to adaptive protection.
Lastly, turn on sign-in and audit logs in the Microsoft Entra admin center. Agent-initiated events are flagged separately from human-initiated ones. This is your baseline for detecting unauthorized access and demonstrating compliance.
- Manage agent identities and inheritable permissions
- Security for AI overview
- Microsoft Entra Agent ID documentation
If you have had a similar experience, please share your learnings in the comments!
Kaitlin Murphy
Senior Director, Identity and Network Access Product Marketing
Additional resources
Learn more about Microsoft Entra
Prevent identity attacks, ensure least privilege access, unify access controls, and improve the experience for users with comprehensive identity and network access solutions across on-premises and clouds.