IBM has announced new identity, access-control and monitoring capabilities designed to help businesses track AI agents, restrict their permissions and review their actions across enterprise systems, according to the company’s October 27 announcement.

The updates include IBM Agent Identity and new capabilities in HashiCorp Vault, which can give agents their own identities and short-lived credentials. 

These controls are designed to limit each agent’s access to the systems and information needed for its assigned task, with the ability to revoke credentials when an agent attempts an unauthorized action.

IBM also announced new capabilities in watsonx Orchestrate to help organizations discover and manage agents built across platforms such as Amazon Bedrock, Microsoft Foundry and Google Gemini. 

The discovery and containment features are in technology preview, while the platform’s agent evaluation capabilities are generally available.

The evaluation tools allow businesses to test agents, models and Model Context Protocol (MCP) servers for quality, accuracy and compliance before deployment, and monitor them in production. IBM’s containment features also include isolated environments and a kill switch for stopping an agent.

Agent Identity and Access Controls Strengthen AI Governance

For chief data officers (CDOs), the announcement addresses a practical problem of not having a complete picture of who owns each agent, what data it can access or what actions it has taken.

These gaps can create risks when agents use persistent credentials or receive broader permissions than their tasks require. A compromised or misconfigured agent could gain access to sensitive business data or act across systems without sufficient oversight.

IBM’s approach reflects a broader shift in AI governance, where organizations need to track agents throughout their lifecycle, assign clear ownership and maintain records of their actions. For data leaders, combining identity controls with access restrictions and audit trails can make it easier to determine what an agent was authorized to do and investigate when something goes wrong.