Join our Newsletter — 33% off our NHI Course
Home› FAQ› Cyber Security› How should security teams stop data leakage when…
Cyber Security

How should security teams stop data leakage when business users build reports on shared semantic models in Microsoft Fabric?

← Back to all FAQ
By NHI Mgmt Group Editorial Team Updated October 31, 2026 Domain: Cyber Security

Security teams should treat the semantic model as the real security boundary, not the report. Restrict who can access the underlying model, review default sharing settings, and remove Q&A features where users do not need free-form querying. If business-led development is unavoidable, map every report to its source model and validate that report-level restrictions still hold in practice.

Why the semantic model is the security boundary in Fabric

In Microsoft Fabric, the report is often just the presentation layer. If multiple business users can build reports from a shared semantic model, leakage usually happens at the model, dataset, or query path, not inside the visual itself. Security teams therefore need to decide access once at the semantic layer and then confirm that report-level permissions do not create a false sense of separation.

The practical implication is that a “safe report” can still expose sensitive fields if the underlying model is broadly accessible, overly permissive, or configured for free-form exploration. That is why model access, default sharing behavior, and query features matter more than the individual report artifact.

When organisations want a deeper pattern for permission-aware data access, the same logic appears in the Permission-Aware RAG Guide: enforce access where retrieval happens, not only where content is displayed.

How leakage happens when business users self-serve reports

Shared semantic models expand the number of people who can discover relationships, infer sensitive data, or reuse fields in unintended ways. Even when a report author does not intend to expose a column directly, default measures, related tables, drill-through paths, and ad hoc querying can reveal more than the published report layout suggests.

That risk is amplified when users are allowed to create many reports from the same model. The security problem becomes one of propagation: one overly broad model becomes many downstream views, and each view can inherit or expose the same weak access decision. If business-led development is the operating model, every report should be traceable to its source model and tested against the same access assumptions.

For a broader Microsoft environment view, the Microsoft 365 Security Guide is useful for understanding how sharing, auditability, and identity controls combine across the tenant.

What to lock down first in Fabric

Start with the model, then work outward. Restrict who can access the underlying semantic model, review whether default sharing is broader than intended, and remove Q&A or free-form querying where users do not need it. Those features are powerful for discovery, but they also widen the chance of accidental disclosure and make it harder to predict what a user can extract.

Business users can still have useful self-service capability without unrestricted exploration. The right control is not to stop reporting altogether, but to constrain the data surface so that people can build within a governed boundary. That usually means limiting model access, validating row and object-level restrictions where they exist, and checking that report consumers cannot bypass the intended view through alternate analysis paths.

Where organisations need a reference point for the cloud responsibility split behind these decisions, Shared Responsibility Models Guide helps frame which access decisions remain the customer’s job even in managed platforms.

Risk and Threat Considerations

Shared semantic models create a classic over-sharing problem: a single mis-scoped model can expose data to many users and many reports at once. The threat is not limited to deliberate abuse. Accidental disclosure, over-broad discovery, and inferred access through self-service querying are enough to create material leakage.

Failure mechanism: Access is granted at the semantic layer more broadly than the business intended, then report authors or consumers reuse that model to surface fields, relationships, or derived values that should not be visible.

Impact: Sensitive data can be exposed across multiple reports, and the organisation may not notice until an audit, a complaint, or an incident review shows that report-level restrictions did not hold in practice.

Standards & Framework Alignment

This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.

NIST SP 800-53 Rev 5 and CIS Controls v8 set the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
NIST SP 800-53 Rev 5AC-6 — Least PrivilegeRestricts access to shared semantic models to only what users need.
AC-3 — Access EnforcementAccess decisions must be enforced at the model layer where data is queried.
AU-6 — Audit Record Review, Analysis, and ReportingTraceability is needed to confirm which reports map to which source models.
Recommendation — Apply least privilege to semantic model access and downstream report consumption. Enforce permissions at the semantic model rather than relying on report boundaries. Review audit evidence to validate report-to-model access and detect over-sharing.
CIS Controls v8CIS-6 — Access Control ManagementShared model access and default sharing require managed access governance.
CIS-16 — Application Software SecurityBusiness-built reporting depends on secure configuration of the analytics layer.
Recommendation — Manage model access centrally and remove unnecessary sharing paths. Harden the analytics configuration that governs semantic model exposure and querying.

Practitioner Guidance

What to verify: Confirm the source model behind every business-built report, then test the report as a low-privilege consumer rather than trusting the authoring view. If a user can reach the data through an alternate path, the control is not working as designed.

Decision rule: If users do not need free-form query capability, disable Q&A and similar discovery features first; if they do need them, treat them as a governed exception with explicit ownership and review. That is usually safer than trying to police every downstream report individually.

What practitioners underestimate: The hardest part is not publishing reports, it is proving that the same restriction model still holds after reuse, copying, and self-service exploration. A report catalog without source-model inventory will miss the real exposure path.

Practitioner takeaway: Treat Fabric semantic models as the enforcement point, because once data is exposed there, report-level controls are usually too late to prevent leakage.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Reviewed and updated by the NHIMG editorial team on October 31, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org