Today, AI agents are no longer just used in demos and assistants. Agents already exist that act autonomously within enterprise systems by completing tasks, calling tools, and continuously executing actions across business processes.

For SAP customers, this development can mean a significant increase in productivity. At the same time, however, this also comes with strict requirements: AI agents for companies must be fundamentally secure, controllable and verifiable.

This is the background for the close technical collaboration between SAP and NVIDIA to integrate the SAP Business AI Platform and NVIDIA OpenShell, a secure open source runtime environment for autonomous AI agents. The goal is not for SAP to “take over” the OpenShell runtime environment. Rather, it's about SAP designing the execution layer for Agentic AI together with NVIDIA, adding security functions and converting it into products.

What benefits collaboration brings for SAP customers

SAP customers benefit from concrete and practical benefits through this collaboration.

It creates the conditions for:

• AI agents that operate within SAP processes without bypassing governance mechanisms
• Security models that align with enterprise IAM and compliance frameworks
• Clear audit trails for agent actions across systems
• a safe transition from pilot projects to productive use

Above all, it ensures that customers do not have to choose between innovation and control. They are not forced to retroactively implement security features or revise their risk models to accommodate the needs of AI agents.

Rather, security and governance are built into the execution model from the start.

The real challenge for companies: building trust in the actions of agents

When AI systems no longer just provide answers to questions but also carry out actions, this brings with it completely new risks.

Agent-based systems can:

• access data in recording systems,
• act across application and data boundaries and
• Carry out actions without having every step checked by a human.

These opportunities pose significant security and governance challenges for all organizations, particularly in regulated industries, when it comes to the security and governance of agent execution. Controls common in the age of chatbots fall short when agents have access to shells, files, networks, credentials, and APIs.

SAP customers know only too well what consequences this can have in practice. The use of AI is only beneficial to them if the AI:

• can be examined and checked for errors,
• is used on the basis of guidelines and
• is trusted by security and compliance teams.

To solve this problem, it is not enough to provide basic infrastructure building blocks or define rules at the application level.

NVIDIA OpenShell as a basis

NVIDIA OpenShell helps solve this problem by securely running autonomous agents in a sandbox environment.

The open source runtime environment OpenShell provides powerful functions. Examples of this are:

• isolated execution environments
• Enforcing policies when accessing the file system and network
• Execution only at the runtime environment level, limiting the consequential damage even if the agent's logic fails

These functions form a foundational layer for the secure execution of autonomous agents. In practice, it is important for companies that this execution layer is aligned with the business context and governance model. Companies expect clear answers to the following questions:

• Which user role authorizes an action?
• Which process context needs to be taken into account?
• What company policies and audit trails apply to actions?

SAP can make a decisive contribution here.

SAP complements OpenShell with business semantics, governance and scaling options

SAP brings the specific requirements of companies into the development of OpenShell.

1. Enterprise requirements for runtime environments

Like no other software provider, SAP is responsible on a large scale for:

• the control of business-critical processes
• supporting companies in regulated industries
• executing millions of transactions per hour

With workloads that are executed in practice by SAP agents, SAP, as part of its collaboration with NVIDIA, provides the operational testing ground with which the powerful OpenShell runtime environment can further develop into a field-tested environment.

In this context, the requirements for the following aspects must also be defined:

• Limits on isolation that correspond to the company's risk models
• Enforce policies in accordance with specific business conditions
• Verifiability that meets customer expectations and legal requirements

2. Joint development of OpenShell functions

SAP provides capacity for the development of the OpenShell open source code base. The focus is on areas that are particularly important for companies:

• Securing the runtime environment
• Policy Modeling
• Integration into corporate identity management
• Hooks for verification and governance

SAP contributes to defining the requirements for the secure execution of agents in the company – not only in theory, but also in productive use.

3. Joule Studio Runtime: from secure execution to enterprise control

While OpenShell enables secure execution, the Joule Studio runtime environment Provides the necessary functions and processes so that agents can be used and controlled by company systems:

• Business-oriented policy semantics (roles, skills, life cycle)
• Identity and access control for companies
• Observability and verifiability all agent behavior
• Implementation and operational governance in all landscapes

This ensures that the autonomous actions of agents are always based on business goals and responsibilities and not just on technical authorizations.

OpenShell's answer is: “Can this action be performed safely?”
The Joule Studio runtime answer is: “Should this action even be carried out?”

Raising the bar for Agentic AI for enterprises

The collaboration between SAP and NVIDIA is not just about integrating technologies. Rather, it represents the shared intention to define the meaning of the term “for companies” in the context of autonomous AI systems.

By bundling

• NVIDIA's innovative runtime environment and security features
• with SAP's ability to translate technologies into products, as well as SAP's governance knowledge and large-scale use of SAP solutions

SAP and NVIDIA are creating an integrated solution for trusted agent execution – a solution that companies can audit, control and deploy securely.

This collaboration provides SAP customers with AI agents that are not only powerful, but also designed to inspire trust in environments where trust is particularly important.

Andre Lamego is SVP and Chief Product Officer for SAP BTP Fabric.