SAP’s Approach to Trusted AI

Understanding how AI works and how data is handled is essential. But in real business environments, trust does not come from understanding alone, it comes from how systems are actually built, secured, and operated.

Many AI solutions define principles such as secure or ethical. The real challenge is implementing these consistently across data, models, and processes.

SAP’s approach focuses on embedding trust directly into the architecture and lifecycle of AI systems, rather than treating it as an additional layer.

SAP delivers AI capabilities through SAP Business Technology Platform (BTP), using services such as SAP AI Core and the generative AI hub.

These services do more than provide access to AI models. They establish a controlled architectural layer that governs how AI is used across the enterprise.

In many AI scenarios outside enterprise environments, applications connect directly to AI models. While this can be fast to implement, it often limits control over data handling, security, and lifecycle management.

SAP takes a different approach. All AI interactions are routed through managed services. This allows organizations to:

• Control how models are accessed and used
• Define which models are available for specific use cases
• Monitor and manage usage across teams and applications

At the same time, this architecture enables integration with business data and processes. AI is not isolated, but connected to enterprise systems in a controlled and structured way.

This architectural layer is what makes it possible to combine:

1. Flexibility (using different models and evolving AI capabilities)
2. Control (through governance and security mechanisms)
3. Scalability (across multiple use cases and business functions)

Security starts with protecting data at every stage of its lifecycle.

Communication with SAP AI Core is protected using modern transport layer security (TLS), ensuring that data in transit is protected against interception. Data stored within SAP-managed environments is protected through encryption at rest, depending on the service and configuration, including backups that are stored securely in separate locations.

Access to AI services is controlled through SAP’s Authorization and Trust Management service (XSUAA). Users and applications authenticate via trusted identity providers, and permissions are assigned based on roles. This allows organizations to enforce least privilege and ensure that access is restricted to what is strictly necessary.

All interactions with AI services are handled through secure APIs. This prevents direct access to models and ensures that every request is authenticated, validated, and traceable. Token-based mechanisms such as JSON Web Tokens (JWT) are used to securely transmit and verify authorization information.

Within SAP BTP environments, applications are typically structured as microservices running in isolated containers. A central application router serves as the secure entry point, managing authentication, session handling, and forwarding requests to backend services. This ensures that access is consistently controlled and monitored.

AI workloads introduce additional considerations beyond standard application security.

In SAP AI Core, workloads run in isolated, tenant-aware environments. Each tenant and resource group is separated through dedicated namespaces, ensuring that workloads cannot access data or processes from other tenants or resource groups.

Containers are executed with enforced security constraints. For example, workloads run in isolated environments with enforced security restrictions, and configurations that introduce elevated privileges are rejected by the platform. This helps reduce the risk of unauthorized system access or misuse.

Customers can deploy their own AI workloads using Docker images. While this provides flexibility, it also introduces responsibility. Customers are responsible for securing their own images and are strongly encouraged to follow secure development practices, such as scanning for vulnerabilities, avoiding sensitive data in images, and maintaining up-to-date dependencies.

Sensitive information used during runtime is handled through secure mechanisms such as secrets, which are created and managed within SAP AI Core and accessed in a controlled way by workloads. Non-sensitive configuration data is handled separately through configurations or labels.

When working with AI, data protection is not handled by technology alone, it is a shared responsibility.

In many enterprise scenarios, SAP provides the infrastructure and services for processing data and acts as a data processor, while customers determine how personal data is collected, used, and managed within their business processes.

This means customers remain responsible for areas such as:

• Ensuring personal data is handled in accordance with applicable regulations
• Obtaining consent from data subjects where required
• Defining retention and deletion policies
• Implementing appropriate governance and compliance processes

At the same time, SAP supports these responsibilities through secure infrastructure, access controls, and data protection measures that help protect data from unauthorized access and unintended exposure.

This reflects a broader principle: secure systems support compliance, but they do not replace governance and organizational responsibility.

Ethical AI requires more than principles, it requires structured processes.

SAP aligns its approach with internationally recognized frameworks, including the UNESCO Recommendation on the Ethics of Artificial Intelligence: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics

These principles are translated into governance practices, where AI use cases are evaluated to identify potential risks and ensure alignment with defined ethical standards. Depending on the level of risk, use cases may require additional review before deployment.

This approach ensures that ethical considerations, such as fairness, transparency, and human oversight, are applied consistently throughout the lifecycle of AI systems.

SAP’s approach to AI is aligned with global standards and regulatory frameworks, ensuring that AI solutions meet recognized requirements for security, governance, and accountability, including standards and guidance such as:

1. EU AI Act
2. ISO/IEC 42001
3. NIST AI Risk Management Framework

These frameworks guide how risks are managed, how systems are governed, and how accountability is ensured.

SAP integrates these standards into a unified approach, rather than treating them as separate requirements. This allows organizations to meet regulatory expectations while maintaining consistency across systems.

Transparency is further supported through the SAP Trust Center, where customers can access information about security, compliance, and data handling practices: https://www.sap.com/about/trust-center.html.

Trust in AI is not established at a single point in time. It is supported across the lifecycle of AI systems.

From defining use cases, to developing and deploying solutions, to operating and monitoring systems, security and governance mechanisms are applied at each stage. Logging and auditing capabilities ensure that key events, such as access to resources or deployment changes, are recorded and traceable.

This lifecycle approach helps ensure that AI systems remain secure, compliant, and reliable as they evolve.

For organizations, adopting AI is not just about what the technology can do, it is about how confidently it can be used.

SAP’s approach provides this confidence by:

• Controlling how models are accessed
• Protecting data throughout its lifecycle
• Embedding governance into the system itself

This allows AI to move beyond experimentation and become a reliable part of real business processes.

Trust in AI depends on more than understanding how AI works. It requires secure architecture, clear governance, and responsible design.

SAP supports trusted AI through:

• Controlled access to AI models through SAP-managed services
• Security measures such as encryption, authentication, and workload isolation
• Clearly defined responsibilities for data protection and governance
• Ethical AI processes and alignment with global standards such as the EU AI Act and ISO/IEC 42001

Together, these measures help organizations use AI in a secure, compliant, and trustworthy way.