IBM puts agent identity alongside controls for credentials and cross-platform discovery

IBM announced general availability of Agent Identity on October 26 and described separate technical previews for cross-platform discovery and containment at TechXchange on October 27.

IBM used its October 27 TechXchange announcement to describe how it wants enterprises to manage AI agents as identities with defined access. IBM had announced general availability of Agent Identity on October 26. At TechXchange it paired that release with updates involving HashiCorp Vault and watsonx Orchestrate, including technical previews for discovering and containing agents. [IBM Agent Identity availability](https://www.ibm.com/new/announcements/ibm-agent-identity-is-now-generally-available-building-trust-for-the-agentic-era) · [TechXchange announcement](https://newsroom.ibm.com/2026-10-27-ibm-introduces-identity-and-control-for-ai-agents%2C-securing-frontier-intelligence-for-the-enterprise)

A named agent still needs a limited credential

Giving an agent its own identity helps an administrator tell which software actor made a request. The next question is what that actor can do. An agent allowed to read an internal record may have no business editing it, and a permission needed for one task may be too broad after the task ends. IBM positions Vault’s credential controls alongside identity so teams can grant narrower access and revoke it when needed. It also describes integration with NVIDIA OpenShell for an execution environment that can use verified identity and short-lived credentials. [TechXchange announcement](https://newsroom.ibm.com/2026-10-27-ibm-introduces-identity-and-control-for-ai-agents%2C-securing-frontier-intelligence-for-the-enterprise)

Availability differs across the product layers

IBM lists cross-platform discovery and sandbox containment as technical previews. It describes evaluation and cost tracking separately. A customer evaluating the product can therefore ask which exact features are available in its environment and which still depend on a preview.

A narrow pilot could start with one agent, one credential and one approved operation. The team can verify that the agent’s actions appear in the relevant logs, that an out-of-scope operation is denied and that revocation works while a session is active. That gives the business evidence about the actual trust boundary, beyond whether an agent appears in a catalogue. [TechXchange announcement](https://newsroom.ibm.com/2026-10-27-ibm-introduces-identity-and-control-for-ai-agents%2C-securing-frontier-intelligence-for-the-enterprise)