ServiceNow builds the trust layer for Autonomous AI with NVIDIA OpenShell

[SamirD]

At Computex this week, NVIDIA announced that the world's leading software platforms are integrating NVIDIA OpenShell as the secure, open-source runtime for autonomous agents. ServiceNow is one of them.

A new class of AI agent has arrived in the enterprise: one that reasons, writes code, executes, and loops until it gets it right. These agents do not answer questions and stop. They take on complex multi-step work, adapt when things do not go as expected, and run autonomously for hours at a time. The question is no longer whether enterprises will deploy them. It’s how to secure and govern them at scale.

 

The enterprise trust layer: secured by NVIDIA, governed by ServiceNow

NVIDIA brings OpenShell, the secure, open-source runtime where autonomous agents execute and where policies are enforced on every action they take. ServiceNow’s AI Control Tower (AICT) is the enterprise AI governance layer: where policy is centralized and authored, where every agent and action taken is logged and observed, and where accountability lives when something goes wrong.

Together, they form the trust layer for autonomous AI in the enterprise – policy is defined centrally in AI Control Tower and enforced at runtime with NVIDIA OpenShell.

NVIDIA OpenShell: an open source, secure runtime for autonomous agents

NVIDIA OpenShell is a secure runtime for developing and deploying  autonomous agents in isolated, sandboxed environments. It is also the enforcement layer that decides, in real time, what each agent is allowed to do. Every file read, command executed, and network call passes through OpenShell first - where agents start with zero permissions and every resource must be explicitly granted. Actions inside policy proceed. Actions outside it are blocked. That’s how autonomous activity stays secure, contained and auditable, even when an agent runs for hours unattended. Because OpenShell is designed to run wherever the agent runs, from PCs to data centers to clouds, the same enforcement layer follows the agent across every environment.

ServiceNow AI Control Tower: enterprise-wide governance for autonomous agents

OpenShell sandboxes every action at the endpoint. But an enterprise running a fleet of agents needs more than endpoint safety. IT and Security departments need to see what every agent in the fleet is doing, push policy that reflects each user's role and permissions, and intervene the moment something goes wrong. That is the work of ServiceNow AI Control Tower.

AICT discovers every autonomous agent active across the enterprise, on every machine, with every skill and tool connection. Teams define and manage role-aware policies for every agent in a single place. These policies are pushed to OpenShell and enforced at runtime. AICT observes LLM calls, tool invocations, and agents’ actions, in real time across the organization. And it gives security leaders the kill switches to terminate any agent instance at any moment.

Project Arc: an autonomous agent the enterprise can trust 

Project Arc is the proof that this model holds together. It is this new class of agent, built by ServiceNow from the ground up with the trust layer at its core. Project Arc thinks, writes code, executes, and adapts when things do not go as expected, completing complex multi-step work across enterprise tools and systems without requiring pre-built workflows. And through Action Fabric, ServiceNow's agentic interface into the platform, Project Arc has the full power of the ServiceNow platform behind every step.

Every critical action Project Arc takes runs inside OpenShell. Every action is governed by AICT, with policies set centrally, behavior monitored continuously, and every file read, command executed, and API call logged. The result is an autonomous agent that an enterprise security leader can audit and approve with confidence.

The trust layer built for what comes next

AI does not sit still. New agents and models arrive continuously, and security and governance models must keep up. From the data centers where models run to wherever the agent acts, ServiceNow and NVIDIA are building the trust layer that lets  a CISO say yes to autonomous AI. Not because they hope it behaves, but because they can see, govern, and secure every action it takes.

Learn more about Project Arc and the ServiceNow and NVIDIA partnership.

[pareekamitk]

hello