AI agents are getting jobs. Now they are getting ID badges, too.
IBM and Nvidia have announced an integration between IBM’s Agent Identity solution and Nvidia’s Open Agent Safety Platform, known as OASP. The goal is simple to describe and hard to pull off: make sure every autonomous AI agent is identified, scoped and watched while it works.
What IBM and Nvidia are actually building
IBM’s Agent Identity entered public preview on September 1, 2026. The integration with Nvidia was highlighted on September 28, 2026, less than a month later.
The basic idea treats each AI agent like a new employee. It gets registered, receives a unique identity, and is handed only the access it needs for a specific task.
That access does not last forever. The system issues scoped, temporary tokens, so permissions expire instead of piling up quietly in the background.
For the credentials themselves, the integration uses HashiCorp Vault. Vault generates dynamic, short-lived credentials for agents, which means a stolen key is only useful for a narrow window.
Red Hat OpenShift handles the infrastructure side. It manages heterogeneous cloud environments, the messy mix of platforms most large companies actually run on.
The shadow agent problem
One of the more interesting pieces involves what the companies describe as shadow agents. These are unknown or undocumented AI entities running inside an organization’s systems.
IBM Identity Protection is designed to discover and monitor all agents, including those shadow agents. You cannot set rules for software you do not know exists, so finding it comes first.
Once an agent is found and registered, its actions can be audited and controlled. That is the core promise: every action traces back to an identity, and every identity traces back to a policy.
Where Nvidia comes in
IBM supplies the identity layer. Nvidia supplies the enforcement muscle underneath it.
Nvidia contributes isolated runtime environments and hardware-level monitoring. Put plainly, agents run in walled-off spaces, and the hardware itself keeps an eye on what they do.
A key component is Nvidia’s Sentry. It can verify an agent’s identity and enforce policies within milliseconds, even when a host is compromised.
More than 100 organizations are reportedly working with Nvidia’s OASP technologies.
Why identity is the new perimeter for AI
The IBM and Nvidia approach is layered. Software-defined identity management decides who an agent is and what it may do. Hardware-based enforcement makes sure those decisions hold at runtime.
The companies also frame the effort around human accountability. Every agent, in this design, ultimately answers to someone, which is a reasonable expectation for software that can move data or trigger transactions without asking first.
The partnership builds on prior work from both companies. It also fits a broader push toward consolidating governance across AI applications rather than bolting security onto each one separately.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.