SAP and NVIDIA Expand Collaboration to Secure Enterprise AI Agents

On May 12, 2026, at SAP Sapphire, SAP and NVIDIA announced an expanded collaboration designed to help enterprises deploy specialized AI agents equipped with rigorous security and governance controls. The announcement occurred during a keynote delivered by SAP CEO Christian Klein, which included a video appearance by NVIDIA founder and CEO Jensen Huang.

The partnership focuses on the transition from AI assistants to autonomous agents within enterprise systems. These agents are increasingly being integrated into core business functions, including procurement, finance, manufacturing and supply chain management, where they access data and execute workflows at scale.

Integrating NVIDIA OpenShell into SAP Business AI

As part of this collaboration, SAP is embedding NVIDIA OpenShell into the SAP Business AI Platform. OpenShell is an open source runtime specifically developed for the secure deployment and development of autonomous AI agents.

Within the SAP Business AI Platform, OpenShell serves as the runtime security layer for all SAP AI agents. This includes custom agents created using Joule Studio, which is SAP’s dedicated environment for managing and building end-to-end enterprise agents.

OpenShell provides several critical security mechanisms to prevent damage in the event of agent logic failure, including:

• Isolated execution environments.
• Infrastructure-level containment.
• Policy enforcement at the network and filesystem layers.

SAP engineers are not merely adopting the technology but are co-designing OpenShell alongside NVIDIA. The two companies are contributing to the open source project with a specific focus on the requirements of production-grade agentic AI, such as auditing and governance hooks, enterprise identity integration, policy modeling, and runtime hardening.

The Trust Equation for Autonomous Agents

The shift toward autonomous agents introduces a different set of trust requirements compared to standard AI assistants. Because autonomous agents can cross application boundaries, operate without a human review at every step, and interact with systems of record, they require strict boundaries and audit trails before they can be deployed in production.

To address these requirements, the collaboration establishes a dual-layer verification process. NVIDIA OpenShell and the Joule Studio runtime work in tandem to ensure agent actions are both safe and authorized.

NVIDIA OpenShell asks: Can this agent action safely execute? Joule Studio runtime — the enterprise control layer within SAP Business AI Platform — asks: Should this action happen at all?

NVIDIA Blog

This architecture is intended to close security gaps that cannot be resolved by application-layer security alone. For agents to be effective, they must understand data boundaries, permissions, roles, and processes, while operating in an execution environment that limits their visibility and the location of their inference runs.

The Application Layer and Economic Value

The strategic importance of this collaboration is framed by Jensen Huang’s description of AI as a five-layer cake, consisting of energy, chips, infrastructure, models, and applications.

According to this framework, the application layer sits at the top and is where AI generates economic value and improves productivity for knowledge workers. Because SAP manages the workflows for finance, supply chain, and procurement, it acts as a catalyst in this layer, ensuring agents operate within identity and process controls.

The collaboration is further informed by NVIDIA’s experience as a customer of SAP, as NVIDIA utilizes SAP for its own logistics, supply chain, and finance operations. This shared context helps both companies define the practical requirements for enterprise-grade governance.

Accelerating Deployment with NemoClaw

To streamline the process for developers, SAP is making NVIDIA NemoClaw available directly within Joule Studio. NemoClaw is a reference blueprint designed for the deployment and development of autonomous agents.

By providing this structured route from the initial build to production, development teams can deploy trusted agents without the need to engineer security scaffolding from the ground up. This is intended to speed up the path to production for SAP customers building custom agents.

the collaboration aims to make AI agents ready for action by ensuring they remain within the governance boundaries required by organizations that house their systems of record within SAP.