Confidential business information can reach AI applications
AI applications and agents can process sensitive business content, so data flows, access, and use cases deserve an intentional security design.
All NLCS services
Microsoft Foundry AI security & governance
Secure Your AI. Protect Your Data. Connect Your Response.
Extend your Microsoft security architecture to AI applications and agents. NLCS helps you implement Microsoft Purview data controls, Microsoft Defender threat protection, and Microsoft Sentinel monitoring around supported Microsoft Foundry workloads.
The business problem
AI workloads need security controls that connect across teams.
As organizations build and operate AI applications, they need to understand how confidential information is used, where security signals are available, and how those signals fit the investigation process their teams already operate.
Visibility depends on the supported workload and configuration
Organizations need to understand what interactions, classifications, audit events, and policy outcomes are available in their selected environment.
Data protection and workload security solve different problems
Purview-oriented data controls and Defender for Cloud workload threat protection are complementary paths that should be coordinated, not conflated.
Security teams need relevant signals in their operating workflow
Available AI, identity, endpoint, and data-protection signals can be connected to an investigation path when supported integrations are configured.
How the Microsoft architecture fits together
Coordinate data security, workload defense, and security operations.
Foundry is the platform where teams build and run AI applications and agents. Purview, Defender for Cloud, Defender XDR, and Sentinel each play different roles. NLCS validates which supported paths apply to the target workload rather than assuming a universal integration.
Illustrative architecture
AI data protection and security signal paths
Roles and paths are confirmed for the scoped workload and target environment.
AI workload
Microsoft Foundry
Build and run AI applications and agents.
Governed AI interaction context
Data controls
Microsoft Purview
Govern supported AI interactions and apply data security and compliance controls.
Threat protection context
Workload security
Defender for Cloud
Detect threats targeting supported AI workloads.
Data policy signal path
01
Purview DLP alerts
Supported alerts and incidents.
Supported integration
02
Defender XDR
Integration path for supported DLP alerts and incidents.
Connector or incident integration
03
Sentinel
Correlate available signals and investigate incidents.
AI workload threat path
01
Defender for Cloud alerts
Available workload threat alerts.
Supported connector
02
Sentinel
Investigate incidents and orchestrate configured response workflows.
Additional input
Identity signals
Configured identity telemetry.
Additional input
Endpoint signals
Configured endpoint telemetry.
Available signals
Operations
Microsoft Sentinel
Correlate available signals, investigate incidents, and orchestrate configured response workflows.
Capabilities depend on workload, configuration, licensing, authentication context, permissions, retention, connectors, and supported integrations. This illustration does not represent every prompt, response, Purview activity, or policy event as automatically available in Sentinel.
What NLCS delivers
Services available for discovery and scoping.
The selected workstream depends on the applications, agents, authentication approach, data flows, subscriptions, licensing, and operational ownership in scope. Service cards describe possible deliverables, not a fixed package or completed customer results.
01
AI Security & Governance Assessment
- Inventory in-scope AI applications, agents, identities, and data flows.
- Identify sensitive-data exposure and governance gaps.
- Deliver a prioritized implementation roadmap.
02
Purview Integration & Data Protection
- Configure supported Foundry–Purview integration.
- Establish sensitive-data visibility and appropriate policies.
- Validate supported audit and compliance scenarios.
- Coordinate application or API integration where enforcement requires it.
03
AI Workload Threat Protection
- Configure applicable Defender for Cloud protections.
- Review access controls and workload security settings.
- Validate available threat alerts and escalation paths.
04
Sentinel Monitoring & Response
- Configure relevant connectors and telemetry sources.
- Develop agreed detection and investigation use cases.
- Create dashboards and response playbooks where appropriate.
- Validate alert routing and incident handling.
05
Ongoing Governance & Optimization
- Review policy effectiveness and integration health.
- Tune detections and reduce unnecessary alert noise.
- Review changes to AI applications and data access.
- Maintain operational documentation and handoff procedures.
AI security assessment checklist
Start the assessment with the architecture and operating reality in view.
The assessment defines what is in scope, confirms the available Microsoft capabilities, and identifies the right implementation and operating decisions before changes are recommended.
Illustrative scenario
From Sensitive AI Interaction to Security Investigation.
An employee submits sensitive contract information to an internal Foundry application. Supported Purview controls provide visibility and, where correctly implemented, policy enforcement. Relevant alerts and connected identity or endpoint evidence support investigation in Sentinel.
Data policy violation
A policy outcome concerns sensitive-data handling. Whether an alert or enforcement result is available depends on the supported configuration and application context.
Attack against an AI workload
A workload-security signal, such as an available threat alert, is a distinct security event from a data-policy outcome.
SOC investigation and response
Configured Sentinel connectors, correlations, investigation workflows, and response playbooks can help the customer’s security team handle relevant signals.
Native setup and developer integration
Start with the supported path. Integrate application controls when the use case requires it.
Native Foundry–Purview configuration can support documented governance outcomes such as audit and AI interaction visibility. For custom application enforcement or other application-level data-control scenarios, the scope can include developer or API integration alongside the client’s engineering team.
01
Assess
Review existing Microsoft security configurations, data risks, and operational requirements.
02
Architect
Design security controls, integrations, evidence paths, and investigation workflows.
03
Implement
Configure approved Microsoft security capabilities and integration components.
04
Validate
Test configurations and document expected versus observed results.
05
Operationalize
Establish investigation procedures, evidence collection, and security-team handoffs.
06
Optimize
Improve control effectiveness, detection quality, and configuration consistency.
The appropriate design depends on the model endpoint, authentication context, user identity, data source, supported features, and policy objective.
Engagement options
Start at the stage that matches the decision in front of you.
Scope, capacity, commercial terms, and any engineering changes are agreed after discovery. Engagements can complement an existing development team, SOC, compliance function, or Microsoft platform team.
Assess
A defined review of AI workloads, sensitive data flows, and security gaps.
Implement
Configuration and validation of agreed data protection, threat protection, and monitoring controls.
Optimize
Recurring governance reviews, tuning, and operational improvement.
Questions buyers ask
Clear scope creates a more useful AI security engagement.
Do we need Microsoft Foundry to use Purview and Sentinel?
No. This service extends the architecture to supported Microsoft Foundry AI workloads. Purview and Sentinel can be used in other Microsoft security and data-protection scenarios.
Does enabling Purview automatically block sensitive AI prompts?
No. Enforcement depends on supported capabilities, policy configuration, user context, and any required application or API integration.
Does Sentinel store every AI prompt and response automatically?
No. Available data depends on configured integrations, telemetry sources, permissions, retention settings, and the supported workload.
Can NLCS work with our existing development team or SOC?
Yes. NLCS can complement existing teams with security architecture, implementation, validation, investigation workflow design, and operational handoff.
Does this make our AI application compliant?
Technical controls can support compliance objectives. Compliance depends on the organization’s broader requirements, evidence, governance, operating practices, and applicable assessment process.
Microsoft support, licensing, and feature status
Validate the actual workload before committing the control.
Visibility is not the same as enforcement
Native Foundry–Purview integration has documented governance paths. Active Data Security Policy enforcement requires supported Microsoft Entra user context; other authentication paths can provide visibility without policy enforcement. Custom applications may require Purview APIs or Agent Framework middleware.
Read official documentation for Visibility is not the same as enforcement (opens in a new tab)Agent, isolation, licensing, and cloud gates
Microsoft’s reviewed agent and Defender-onboarding pages differ on agent coverage, so it must be verified for the actual endpoint and tenant. The documented integration does not support network isolation. Data Security Policies require appropriate pay-as-you-go or Agent 365 entitlement; government-cloud integration parity is not established.
Read official documentation for Agent, isolation, licensing, and cloud gates (opens in a new tab)Documentation reviewed November 5, 2026. Microsoft availability and preview status can change; scope approval includes a fresh tenant-specific check. These are documented capabilities and constraints, not claims of completed NLCS customer deployments.
Start with the AI workload in front of you
Discuss Your AI Security Needs.
Bring the AI use case, Microsoft environment, data sensitivity, available security controls, and decision you need to make. NLCS will help frame an appropriate security architecture and delivery path.
Schedule an AI Security Assessment