IBM has introduced a control layer for enterprise agents that combines discovery, verified identity, credential restrictions, testing and shutdown controls. HashiCorp Vault’s Agent Identity is designed to give each agent a governed identity and enforce least-privilege access, including real-time credential revocation. IBM watsonx Orchestrate is being extended to discover agents built on Amazon Bedrock, Microsoft Foundry and Google Gemini.
The company says discovery and containment are initially available as technology previews. The containment capability runs agents in isolated sandboxes and includes a kill switch, while evaluation tools test quality, accuracy and compliance before deployment and monitor production behaviour and cost. IBM also says the discovery layer can locate models and Model Context Protocol servers, broadening the inventory beyond conventional applications.
The announcement is useful because it frames agent governance as an identity and authority problem, not only a model-safety problem. A registry can show that an agent exists, but it cannot determine whether the agent should be allowed to retrieve a customer record, invoke a payment tool or pass credentials to another process. That requires policy outside the model, strong ownership and an auditable chain from identity to action. Enterprises assessing the tooling should verify how the previews handle false discovery, revocation latency and evidence retention across third-party platforms.
