NVIDIA and SAP Collaboration: Building Trustworthy AI Agents for Enterprise

In a major step toward enterprise-grade AI, NVIDIA and SAP have expanded their collaboration to embed security and governance controls into specialized AI agents. Announced at SAP Sapphire 2025, the partnership integrates NVIDIA's OpenShell runtime into SAP's Business AI Platform, enabling autonomous agents to operate safely within finance, procurement, supply chain, and manufacturing systems. This Q&A explores the details, implications, and benefits of this initiative.

What did NVIDIA and SAP announce at SAP Sapphire 2025?

During SAP Sapphire 2025, SAP CEO Christian Klein featured a video message from NVIDIA founder Jensen Huang announcing an expanded collaboration. The core announcement is the embedding of NVIDIA OpenShell, an open-source runtime for secure AI agent development and deployment, directly into the SAP Business AI Platform. This means all SAP AI agents—including custom agents built in Joule Studio—will operate within OpenShell's security framework. Additionally, SAP engineers are co-developing OpenShell alongside NVIDIA, contributing improvements back to the open-source project. The goal is to make enterprise AI agents both powerful and trustworthy, capable of handling critical business workflows without compromising security or governance.

How does OpenShell ensure security for autonomous AI agents?

OpenShell provides a multi-layered security approach tailored for autonomous agents. It creates isolated execution environments that prevent agents from interfering with each other or with underlying systems. At the filesystem and network layers, policy enforcement restricts what an agent can read, write, or communicate with. Infrastructure-level containment acts as a safety net, guarding against damage if agent logic fails or behaves unexpectedly. Within SAP's platform, OpenShell becomes the runtime security layer for all AI agents, enforcing boundaries, permissions, and audit trails. This addresses the critical trust equation: as agents move from simple assistants to autonomous decision-makers, they need robust guardrails to safely interact with enterprise systems of record, cross application boundaries, and operate without human review at every step.

Why is the application layer critical for AI agent adoption?

NVIDIA CEO Jensen Huang has described AI as a five-layer cake: energy, chips, infrastructure, models, and applications. The top application layer is where AI generates real economic value and boosts knowledge worker productivity. SAP is a global leader in enterprise applications, running core processes like finance, procurement, supply chain, and manufacturing. For AI agents to be trusted in these domains, they must operate within existing policy, identity, and process controls. SAP's position at the heart of enterprise operations makes it a key catalyst for agentic AI adoption. Business agents need to understand roles, permissions, and data boundaries; they also require an execution environment that limits what they can see and do. OpenShell, integrated into SAP's platform, provides precisely that environment, ensuring agents stay within safe operational parameters while delivering efficiency gains.

What enterprise governance requirements are being addressed by the collaboration?

The shift from AI assistants to autonomous agents dramatically changes the trust equation. An agent that can touch systems of record, cross application boundaries, and act without step-by-step review needs clear boundaries, policy enforcement, and a complete audit trail before it can be deployed in production. The SAP-NVIDIA collaboration is directly tackling these requirements by co-developing OpenShell's open-source codebase. Focus areas include runtime hardening to withstand attacks, policy modeling to define what agents can and cannot do, enterprise identity integration to link agents with user roles and permissions, and auditing and governance hooks for compliance and traceability. Together, SAP and NVIDIA are building the foundational governance framework that enterprises need to deploy agentic AI confidently and responsibly.

How does NVIDIA's own experience as an SAP customer influence this collaboration?

NVIDIA brings a unique perspective to the table—it has been an SAP customer for years, running its own finance, supply chain, and logistics operations on SAP systems. This gives both companies a shared, practical understanding of what enterprise-grade governance requires in real-world scenarios. NVIDIA knows firsthand the complexities of integrating AI into mission-critical business processes where security, compliance, and reliability are non-negotiable. This customer lens informs the design of OpenShell's enterprise features, ensuring they meet the stringent demands of large organizations. By working side by side with SAP engineers, NVIDIA contributes insights that bridge the gap between cutting-edge AI capabilities and the operational realities of global enterprises, ultimately making the solution more robust and trustworthy.

What is the significance of the shift from AI assistants to autonomous agents?

Traditional AI assistants typically respond to queries and require human approval before taking actions. Autonomous agents, however, can independently execute tasks, trigger workflows, and make decisions across different enterprise systems. While this promises dramatic productivity gains, it also introduces new risks. An agent acting without oversight could inadvertently modify critical data or breach process boundaries. The SAP-NVIDIA collaboration addresses this by embedding security and governance directly into the agent's runtime environment. As outlined in OpenShell's security features, isolated execution and policy enforcement ensure agents operate within approved limits. This shift matters because it enables enterprises to safely automate complex, multi-step business processes—from procurement to manufacturing—while maintaining control, auditability, and compliance. Trust becomes the foundation for scaling agentic AI in production environments.