Clear IAM guidelines for AI use are essential

Artificial intelligence (AI) is fundamentally transforming how companies handle data, processes, and decisions. AI also unlocks enormous potential within SAP system landscapes – whether through automated analytics in finance, intelligent assistance systems in procurement, or context-based decision support in HR. However, these new technological possibilities present a key challenge for enterprise security: It is essential to ensure that AI only accesses data that it is actually permitted to see within the respective usage context.

Enablement Package for SAP Joule: Make your IAM ready for AI use

AI must not become a security risk

SAP security officers, IAM managers, and CIOs must be aware: Without robust Identity & Access Management (IAM), the use of AI quickly becomes a security and compliance risk. The introduction of AI is therefore not merely an innovation project but also requires accompanying measures for governance and security. In traditional IAM models, users are identified, authenticated, and assigned roles. These roles clearly define which transactions may be executed and which data may be viewed. But what happens when AI acts as an additional entity in the system landscape? Such non-human identities must also be managed, monitored, and controlled.

The key question is: Does the AI act on its own authorization or in the context of the user making the request? This is precisely what determines whether a company retains its data sovereignty or unwittingly erodes it. Let’s imagine a sales representative asks an AI assistant for a revenue forecast for a specific customer. Does the AI access all data available within the company—including sensitive HR or compliance information—simply because it is technically capable of doing so? Or is access limited to the data that the sales representative is authorized to view based on their role?

The principle of context-based authorization

A robust IAM for AI follows a clear principle: The AI must never know more than the human on whose behalf it is acting. It is not a privileged superuser, but a delegated agent. The AI operates exclusively within the security context of the authenticated user. Technically, this means that identity and authorizations – for example, via token-based methods or OAuth delegation – are passed on to the AI component. Thus, the AI does not process information based on its own global set of authorizations, but rather within the boundaries of the respective business role.

Especially in hybrid architectures, where SAP systems are combined with cloud services and AI platforms from providers such as Microsoft or OpenAI, this seamless transfer of identity and role information is essential. Otherwise, uncontrolled data flows across system boundaries will result.

The focus is on data access

There is another aspect to consider: While traditional authorization concepts are highly function-oriented—that is, focused on transactions and applications—artificial intelligence is shifting the focus more toward data access. The question is less whether a user is permitted to execute a specific transaction, and more about which data may be analyzed, combined, and potentially re-contextualized, and to what extent.

This is where modern approaches such as attribute-based access control (ABAC) are gaining importance. Instead of static roles, dynamic attributes determine access: role, department, data sensitivity level, geographic context, or even the specific use case. For SAP and IAM managers, this means that existing role models must be critically reviewed and evolved.

AI activities must be traceable

In addition to access control itself, the traceability of AI activities is a fundamental feature of a secure IAM framework. Regulatory requirements such as the EU General Data Protection Regulation (GDPR), ISO 27001, NIS 2, and other industry-specific standards require transparency regarding who was authorized to access which data and when. This requirement applies not only to human users.

When an AI performs an analysis or generates a recommendation, it must be clearly documented which authorizations the user had at the time of their query for which data (sources) that were incorporated into the query’s result. Without consistent logging of authorization assignments, dangerous audit gaps arise that can have significant legal and financial consequences in the event of an incident.

A professional IAM for artificial intelligence therefore integrates logging, monitoring, and recertification processes from the outset. This also includes regularly verifying whether roles and authorizations still align with actual needs – especially in dynamic organizational structures.

Zero trust as a guiding principle for AI

The integration of artificial intelligence into business processes should consistently follow the zero trust principle. This means: no access without verification, no implicit assumptions of trust – not even for internal systems or seemingly harmless assistant functions. Every AI request is authenticated and authorized. Access rights are granted according to the least-privilege principle and reviewed regularly. Particularly sensitive data sets should also be segmented and classified so that AI models can only access approved data areas.

For SAP environments with legacy authorization structures, this approach can certainly pose a challenge. At the same time, however, the introduction of AI also offers an opportunity to streamline historically evolved role models and strategically realign IAM.

SAP Joule: AI assistant with integrated authorization concept

A concrete example of an IAM-compliant AI approach in the SAP environment is SAP Joule – SAP’s generative AI assistant, which is already integrated into SAP S/4HANA Cloud, SAP SuccessFactors, and other SAP applications. SAP Joule consistently follows the principle of context-based authorization: The copilot operates exclusively within the authorization context of the logged-in user. Technically, this is implemented via OAuth 2.0 and principal propagation – SAP Joule does not receive any independent system rights beyond those of the user, but rather recognizes the user’s role(s) and authorizations. Accordingly, the user is unable to access information or execute business processes for which they do not have authorization. In this respect, SAP Joule demonstrates how AI functionality and robust IAM design can be reconciled – without compromising data security or compliance.

Conclusion: IAM as an enabler for trustworthy AI

For CIOs, Identity & Access Management in the context of artificial intelligence is far more than a technical detail. It is the prerequisite for ensuring that AI can be deployed in a scalable, audit-proof, and trustworthy manner. Only when clear identities, well-defined role models, and transparent access controls are established, the necessary trust is built – among business units, compliance officers, and, not least, customers and partners.

The introduction of AI should therefore always go hand in hand with a corresponding IAM strategy. Companies that implement AI in isolation from their existing security and governance structures risk losing control over their most sensitive asset: their data.

The key message is therefore: No AI without clear IAM guidelines. Those who establish context-based access controls, modern authorization models, and full auditability today lay the foundation for secure and sustainable AI use within the enterprise – especially in the demanding and sensitive SAP environment. IAM thus evolves from an operational authorization manager to a strategic architect of a trustworthy AI future.