IBM Adds Identity and Access Controls for Enterprise AI Agents

IBM on Tuesday announced controls for enterprise AI agents that assign identities, restrict credentials and evaluate behavior, while its agent discovery and sandbox containment tools remain in preview.

At TechXchange 2026, the company listed IBM Agent Identity, HashiCorp Vault 2.2 and Evaluate in watsonx Orchestrate as generally available. Agent Identity had reached general availability on Oct. 26, one day before IBM’s announcement.

IBM Agent Identity stores an agent’s designated owner, current lifecycle phase and permissions delegated for work across systems. Vault 2.2 can give an agent temporary credentials limited to the minimum required access and end its session after an unauthorized action.

Evaluate checks agents, models and Model Context Protocol (MCP) servers against quality, accuracy and compliance criteria before deployment, then supports production monitoring. Its cost tracking shows model and token consumption.

The preview tools Discover and Contain cover agent visibility and runtime response. Discover searches for agents created on Amazon Bedrock, Microsoft Foundry or Google Gemini; Contain places agents in isolated sandbox environments and offers a kill switch.

The new controls follow IBM Field CTO Tyler Lynch’s earlier discussion of agent safeguards, which framed token monitoring and API-key revocation as ideas for further work.

“The winners in enterprise AI are changing how their business runs, not simply adding more AI to it,” said Dinesh Nirmal, senior vice president of IBM Software.

TechXchange 2026 runs through Oct. 29 in Atlanta.