IBM and Nvidia integrate AI agent identity with runtime access controls

IBM and Nvidia have announced an integration between IBM’s Agent Identity solution and Nvidia’s Open Agent Safety Platform (OASP) to enhance security for enterprise AI agents. The collaboration aims to ensure that autonomous AI agents are identified, scoped, and monitored during operation.

The system treats AI agents similarly to employees by assigning them unique identities and limited, temporary access. Using HashiCorp Vault, the integration generates dynamic, short-lived credentials to prevent unauthorized data access. IBM’s Agent Identity manages the identity and permission layer, while Nvidia provides enforcement through hardware-level monitoring and isolated runtime environments. Specifically, Nvidia’s OpenShell software and Sentry—a hardware-layer design for BlueField-4 DPUs—can isolate agents that attempt to exceed their set boundaries within milliseconds.

Additionally, the partnership addresses the issue of ‘shadow agents’—undocumented AI entities running within an organization. IBM Identity Protection is designed to discover and monitor these unknown agents, ensuring all AI activity can be audited and traced back to a specific identity and policy.