Building Trust in Enterprise AI Agents: How SAP and NVIDIA Collaborate on Secure, Governed Automation

As artificial intelligence evolves from simple assistants to autonomous agents that can touch systems of record, cross application boundaries, and execute workflows without human oversight, enterprises face a new challenge: how to trust these agents. SAP and NVIDIA have expanded their collaboration to address this very issue, embedding security and governance controls directly into the runtime environment of AI agents. By integrating NVIDIA's OpenShell into the SAP Business AI Platform and jointly developing its open-source capabilities, the two companies are creating a foundation for trustworthy, production-ready agentic AI. Below, we answer key questions about this partnership and its implications for enterprise automation.

What are specialized AI agents and why do they need trust?

Specialized AI agents are advanced software programs designed to perform complex tasks across enterprise domains such as finance, procurement, supply chain, and manufacturing. Unlike simple chatbots or assistants that only provide information, these agents can interact directly with enterprise systems—making decisions, accessing data, and executing workflows at scale. This autonomy introduces a critical trust equation: an agent that operates without constant human review must have clear boundaries, policy enforcement, and an audit trail to prevent unintended actions or security breaches. Without trust, enterprises cannot safely deploy these agents into production where they touch sensitive systems of record. The collaboration between SAP and NVIDIA directly tackles this need by embedding security and governance at the runtime level, ensuring agents operate within defined policies and organizational controls.

Source: blogs.nvidia.com

How are SAP and NVIDIA collaborating to secure AI agents?

SAP and NVIDIA are collaborating by integrating NVIDIA's OpenShell—an open-source runtime for securely developing and deploying autonomous AI agents—into the SAP Business AI Platform. This integration makes OpenShell the security layer for all SAP AI agents, including custom agents built in Joule Studio, SAP's environment for building end-to-end enterprise agents. Furthermore, SAP engineers are codeveloping OpenShell alongside NVIDIA, contributing their expertise to the open-source project. This joint effort focuses on hardening the runtime for enterprise use, adding policy modeling, identity integration, and auditing hooks. By working together, both companies ensure that the security and governance features meet the real-world demands of large-scale enterprise operations, drawing on NVIDIA's own experience as a long-time SAP customer running finance, supply chain, and logistics on SAP systems.

What is NVIDIA OpenShell and how does it provide security?

NVIDIA OpenShell is an open-source runtime designed for securely developing and deploying autonomous AI agents. It provides isolated execution environments that limit what an agent can see and do, along with policy enforcement at the filesystem and network layers. These infrastructure-level containment measures guard against damage when agent logic fails or is compromised. Think of it as a secure sandbox: each agent runs in its own confined space where its actions are monitored and restricted according to predefined policies. For example, an agent handling procurement might only be allowed to read specific databases and submit orders within certain budget limits. OpenShell ensures that even if an agent malfunctions or attempts unauthorized actions, the blast radius is contained, and the rest of the enterprise remains safe. This runtime layer is essential for meeting enterprise security and compliance requirements.

Why does the application layer matter for agentic AI?

NVIDIA CEO Jensen Huang famously described AI as a five-layer cake: energy, chips, infrastructure, models, and applications. The top layer—applications—is where AI creates real economic value by driving productivity for knowledge workers. SAP is a global leader in enterprise applications, running critical workflows for finance, procurement, supply chain, and manufacturing. For agentic AI to succeed in business, agents must operate within existing policy, identity, and process controls that only the application layer can provide. This makes SAP's position at the core of enterprise operations a key driver for adoption. Without proper integration at the application layer, agents would lack context about roles, permissions, and data boundaries. By embedding security directly into the SAP Business AI Platform, NVIDIA and SAP ensure that agents respect enterprise governance while still delivering automation benefits.

Building Trust in Enterprise AI Agents: How SAP and NVIDIA Collaborate on Secure, Governed Automation — Source: blogs.nvidia.com

How do SAP and NVIDIA address governance for enterprise agents?

Governance for enterprise agents involves ensuring they understand roles, processes, permissions, and data boundaries. SAP and NVIDIA address this by codeveloping OpenShell to include policy modeling, enterprise identity integration, and auditing and governance hooks. Policy modeling allows organizations to define rules about what agents can access and perform. Identity integration ties agent actions to existing enterprise user roles and permissions, so an agent can only act within the authority granted to its associated users. Auditing hooks provide a complete record of every action an agent takes, enabling compliance teams to review and verify activities. Additionally, the execution environment limits what an agent sees, where inference runs, and what actions are permitted. This multi-layered approach ensures that as agents become more autonomous, they remain under organizational control and can be trusted in production environments.

What specific contributions are SAP engineers making to OpenShell?

SAP engineers are actively contributing to OpenShell's open-source codebase, focusing on features needed for enterprise-grade production deployment. Their key contributions include runtime hardening to improve stability and security under heavy loads, policy modeling to make it easier for organizations to define and update agent constraints, enterprise identity integration to connect OpenShell with existing identity providers like SAP's Identity Management, and auditing and governance hooks that feed into enterprise logging and monitoring systems. By collaborating directly with NVIDIA, SAP ensures that these features meet the practical needs of large enterprises running critical business processes. This joint development also benefits the broader open-source community, as the improvements are contributed back to the project. Both companies share a deep understanding of enterprise requirements, with NVIDIA itself using SAP systems for its own finance and supply chain operations, providing firsthand insight into what governance truly requires.

How does this collaboration help enterprises move from AI assistants to autonomous agents?

The shift from AI assistants—which provide information but don't act independently—to autonomous agents that execute tasks without step-by-step review fundamentally changes the trust equation. Enterprises need confidence that agents won't cause damage or violate policies. The SAP-NVIDIA collaboration directly addresses this by embedding security, policy enforcement, and audit trails into the runtime environment. OpenShell provides the containment and governance needed to let agents operate within systems of record, cross application boundaries, and run without constant supervision. For enterprises, this means they can safely deploy agentic AI to automate complex workflows in finance, supply chain, and manufacturing, gaining productivity while maintaining control and compliance. The partnership ensures that the same governance that applies to human employees also applies to AI agents, making the transition from assistants to autonomous agents both practical and trustworthy.

Tags: