On Oct. 27, 2026, IBM announced enterprise AI-agent identity and control capabilities at TechXchange 2026 in Atlanta. The architecture separates IBM Verify’s identity, authorization, token exchange and audit functions from HashiCorp Vault’s role managing secrets and supplying just-in-time credentials to systems that cannot consume delegated tokens directly.
IBM’s announcement also covered an integration with NVIDIA OpenShell, while the availability of the company’s agent-related features varies by product. IBM had announced IBM Agent Identity as generally available on Oct. 26.
IBM Verify and Vault handle different parts of access
IBM assigns identity and delegated authorization to IBM Verify. HashiCorp Vault protects secrets and brokers short-lived, target-specific credentials for systems that cannot use delegated tokens. That split matters when a service receiving an agent’s request cannot accept the delegated token itself: Vault can provide credentials for that target system.
IBM describes the identity layer as a central registry for agents. It records ownership and metadata, classifies risk, lists approved tool connections and tracks lifecycle status. IBM says its OpenShell integration pairs a protected runtime with verified agent identity and Vault credentials, with the goal of making agent actions traceable and constrained.
A persistent Agent ID does not mean permanent task access
IBM documentation says each registered agent receives a durable Agent ID that persists through credential rotations. The identifier stays with the agent, while the authority used for a task can be short-lived and scoped to that task.
IBM describes OAuth-based delegation, a way to pass authorization between services, as preserving both the human user’s identity and the agent’s role in a request. An access token is a digital credential presented to a service; IBM says these tokens can be short-lived and limited to a task. When a target system cannot accept a delegated token, Vault can supply a separate, target-specific credential.
The onboarding process also separates responsibilities. An Agent Developer creates the agent record and registers its OAuth clients, but cannot change the agent’s status. An Approver can activate the agent but cannot edit its data fields. IBM says suspending an agent disables its linked OAuth clients, so new token requests fail until the agent is reactivated.
IBM lists different availability stages by feature
IBM’s Oct. 27 announcement lists IBM Agent Identity, HashiCorp Vault 2.2 and watsonx Orchestrate Evaluate as generally available. IBM had announced IBM Agent Identity’s general availability the day before, on Oct. 26.
Vault 2.2 includes protected agent accounts, automatic credential renewal and post-quantum encryption, according to IBM. IBM says watsonx Orchestrate Evaluate can assess agent quality, accuracy and compliance before deployment, monitor agents in production, and track model and token use.
IBM lists Secure Hub in public preview. The company describes it as a way to view Vault environments centrally. watsonx Orchestrate Discover and Contain are technology previews: Discover connects with Amazon Bedrock, Microsoft Foundry and Google Gemini to find externally built agents, models and MCP servers; Contain includes isolated sandboxes and a kill switch.
