The image generated by AIYXL using AI
An employee asks an AI agent to investigate a delayed order. Reading its status may be appropriate; changing payment details requires different authority.
That distinction sits at the centre of IBM’s October 27 announcements: identify the agent, define its permissions and retain a way to withdraw them.
Identity needs an owner
IBM says Agent Identity records an agent’s owner, lifecycle status and delegated authority. HashiCorp Vault supplies credential controls, including short-lived access and revocation.
The announcement also separates capabilities by release status. Agent Identity, Vault 2.2 and watsonx Orchestrate evaluation are generally available. Secure Hub is in public preview; Orchestrate discovery and containment are in technology preview.
The proposed management workflow combines finding agents across external platforms, isolating their execution and assessing quality, compliance and cost. Its practical scope therefore depends on which components an organisation can deploy today.
A useful boundary for business tasks
AIYXL’s engineering reading is that an agent should receive authority appropriate to its current task.
In the order example, an investigation could begin with read access. Updating a shipping instruction could require additional permission and a recorded approval. Completion should end the temporary access.
That approach also makes incidents easier to examine. A reviewer needs to connect the initiating request, agent identity, credential and resulting action. A shared account obscures those relationships.
IBM’s announcement establishes product capabilities and availability, rather than independently measured security outcomes. A deployment still needs to demonstrate that permissions expire, revoked credentials stop working and action records remain traceable across connected systems.
Sources: IBM Newsroom