Your AI Agent Should Not Be You: Why Agent Identity Is the Missing Governance Layer for Enterprise AI

Ask a security team how many AI agents run in the enterprise and which identity each one uses, and the answer is often an estimate. 

Business units, platform teams and software vendors all create agents, and many of those agents authenticate with shared service accounts, static API keys or the personal credentials of whoever built them. 

IBM describes this pattern directly, and it leaves auditors unable to tell whether a person or an agent performed an action. Agent identity closes that gap, and it is the governance layer most enterprise AI programs still lack.

IBM watsonx Orchestrate now offers Agent Identity alongside expanded third-party agent management, giving IT teams a way to build multi-agent ecosystems that auditors can trust.

Why Borrowed Identities Break Governance

IBM’s September 2026 announcements name three risks that arise when an agent inherits a user’s identity: limited traceability, overprivileged access and limited authorization control. 

An agent that lists open positions on behalf of an administrator receives the administrator’s full permissions, although read access would cover the task. 

Shared service accounts add another layer of difficulty, since logs point to an account that many agents use. Static keys embedded in agent configurations extend the exposure further.

How Agent Identity Works in watsonx Orchestrate

Agent Identity assigns each agent a unique, verifiable enterprise identity, kept separate from its creator, owner and end user. 

The design integrates with your existing identity provider (IdP), which remains the source of truth. The private preview, released 15 September 2026, supports IBM Verify and Microsoft Entra, with other IdPs and workload-identity patterns on the roadmap.

An administrator first establishes trust between the watsonx Orchestrate tenant and the IdP. Agents built in Orchestrate, and external agents registered with it, then link to identities in the enterprise identity system. 

At runtime, Orchestrate obtains an identity token for the agent and checks that the agent is recognized and authorized for the requesting user. When the agent calls a protected tool, it receives a short-lived on-behalf-of token scoped to that task.

The execution context preserves two roles: the subject, meaning the user the work is done for, and the actor, meaning the agent doing it. Audit records can therefore trace an action through the originating user, the agent, the authorization decision, the credential issued and the downstream tool invoked.

Lifecycle events stay synchronized with the IdP, and disabling an agent’s identity there can block the agent from running, which gives security teams a direct control over shadow agents. IBM also plans to integrate HashiCorp Vault’s agentic IAM capabilities for short-lived credentials to downstream systems.

One Control Plane for Agents Built Anywhere

The watsonx Orchestrate control plane keeps a single inventory of agents from connected platforms alongside native agents. IBM’s October 2026 update lists Amazon AgentCore support through the AI Gateway, plus discovery and import previews for Microsoft Foundry and Google Gemini Enterprise Agent Platform. 

Earlier IBM announcements list LangGraph, Langflow and A2A-protocol agents among supported frameworks. Policies set at the gateway apply to every connected agent wherever it runs, and teams can find overlapping agents built on different platforms and retire the extras.

What IT Teams Should Plan For

Agent Identity is in private preview, so production plans should include onboarding through the watsonx Orchestrate team and a staged rollout. 

Useful early work includes inventorying existing agents and the credentials they use, confirming which IdP will be the source of truth, and defining least-privilege scopes for each agent task. Agents also reach into hybrid estates, including workloads on Red Hat platforms, so scope definitions should follow the systems each agent touches.

This work combines identity and access management, platform engineering and AI governance, a mix of skills that is hard to find. 

Organizations that hire IT talent for these roles face long searches, and many choose to recruit IT talent through a partner. ASB Resources offers IT talent headhunting and staffing alongside our delivery consultants, so your program keeps moving while the right people join.

How ASB Resources Helps

We help organizations establish the identity, access and policy foundations that make multi-agent AI auditable, compliant and trustworthy, regardless of where agents were built. Our engagements cover:

• Agent estate discovery and credential inventory
• IdP integration design with IBM Verify or Microsoft Entra
• Least-privilege scope and on-behalf-of policy design
• Gateway and control plane policies for native and third-party agents
• Audit and lifecycle processes that tie every agent action to a user and an owner

Could you tell right now which agent performed the last action in your most sensitive system?

Let the experts at ASB Resources help you design the agent identity, access and policy foundations your watsonx Orchestrate ecosystem needs. Schedule a call with one of our experts today!