✅ Correct Answers

Deploying a Copilot Studio agent across multiple channels while maintaining security and compliance requires establishing foundational controls before publishing. The case study explicitly states that disclaimers must be applied consistently across all generative responses and that manual edits to individual topics must be avoided. This requirement directly points to the need for prompt modifications configured at the system level.

Prompt modifications work by injecting system-level instructions into the generative AI model that processes user queries. These instructions persist across all topics, channels, and conversations without requiring per-instance configuration. When properly configured, prompt modifications can enforce brand voice and tone, include required legal disclaimers, define content boundaries, specify refusal scenarios for inappropriate requests, and establish response formatting guidelines. For Blue Yonder Airlines, this means every generative response will automatically include appropriate disclaimers and maintain consistent tone whether the interaction occurs on the website, mobile app, or Microsoft Teams.

The second essential pre-deployment action is configuring Power Platform DLP policies. These policies operate at the environment level and control which connectors can be used by all resources within that environment including agents and flows. Blue Yonder's agent needs to integrate with multiple systems: an internal reservation system through a custom connector, external flight status APIs, SharePoint knowledge bases, Dynamics 365 loyalty data, and partner travel advisory services. Without properly configured DLP policies, there's risk that unauthorized connectors could be used, potentially creating data exfiltration paths or compliance violations.

DLP policies work by classifying connectors into groups such as Business, Non-Business, and Blocked. Policies then define which combinations of groups can be used together in a single resource. For example, Blue Yonder might place their internal reservation system custom connector and Dynamics 365 connector in the Business group, while blocking social media and consumer cloud storage connectors entirely. This ensures that customer reservation data cannot be transmitted to unauthorized external services even if someone later attempts to add such functionality.

The question specifically asks what must be done before making the agent available on both channels. This timing is critical because implementing these controls after deployment creates a security window where the agent operates without proper governance. Channel-level configurations are insufficient because they don't provide consistent cross-channel enforcement, and post-deployment individual channel configuration creates deployment complexity and potential inconsistency. System-level controls implemented before any channel is activated provide comprehensive protection from the first user interaction.

The incorrect options represent common anti-patterns: manually editing topics doesn't scale and violates the stated requirement, relying on channel-level settings creates inconsistent protection, and enabling controls after publishing exposes the organization to risk during the deployment window.

Option A (Configure prompt modifications to enforce tone, disclaimers, and refusal behavior at the system level) directly addresses the explicit requirement that disclaimers must be applied consistently across all generative responses without manual edits to individual topics. Prompt modifications are the only mechanism in Copilot Studio that can achieve this requirement, as they provide centralized control over generative AI behavior that applies automatically to all topics and channels.

Option D (Configure Power Platform DLP policies to restrict unauthorized data connectors) satisfies the technical requirement that Power Platform DLP policies must be enforced to block unauthorized data flows. Given that Blue Yonder's agent integrates with multiple sensitive data sources across internal and external systems, DLP policies provide the essential security boundary that prevents unauthorized data access or transmission regardless of which channel is being used.

Both actions must be completed before deployment because they establish the foundational security and compliance framework. Prompt modifications ensure that every AI-generated response meets compliance standards, while DLP policies ensure that every data access follows security policies. Together, they provide defense-in-depth protection that operates consistently across web, mobile, and Teams channels. These are prerequisites rather than post-deployment enhancements because deploying without them would create a period of non-compliant operation.

Option B: Manually add disclaimers to each topic before publishing

This approach directly violates the stated requirement that "Disclaimers must be applied consistently across all generative responses. Manual edits to individual topics must be avoided." Manual per-topic edits create unsustainable maintenance overhead, introduce the risk of inconsistency where some topics might lack proper disclaimers, and fail to scale as new topics are added. Additionally, manually added text in topics doesn't affect generative answers created by the AI model from knowledge sources, meaning disclaimers wouldn't appear on many responses even if manually added to topics.

Option C: Embed the web channel and then rely on channel-level settings to enforce content moderation

Content moderation must be enabled at the platform level, not configured separately per channel. Relying on channel-level settings creates the risk that protection might not be consistently applied across all deployment channels. Additionally, embedding a channel and then configuring settings afterward means the agent could be briefly accessible without proper moderation controls. The case study requires that Responsible AI content moderation filters be enabled, which is a platform-level configuration, not a channel-specific setting.

Option E: Publish the agent and then enable Responsible AI filters individually for each channel

This approach creates a deployment window where the agent operates without essential safety controls, violating compliance requirements. Responsible AI filters should be enabled before publication, not after. Furthermore, Responsible AI content moderation is not configured individually per channel but rather enabled at the platform level where it automatically applies to all channels. This option represents a fundamentally incorrect understanding of when and how content moderation is implemented.

Prompt Modifications: System-level instructions injected into the generative AI model that influence agent behavior, tone, disclaimer inclusion, and response characteristics across all topics and channels without requiring per-topic configuration.

Power Platform DLP Policies: Data Loss Prevention policies configured at the environment level that control which connectors and data sources can be used, preventing unauthorized data flows and enforcing security boundaries.

System-Level Configuration: Settings and controls applied at the platform or environment level that automatically affect all agents, topics, channels, and conversations, ensuring consistent enforcement.

Multi-Channel Deployment: The process of making a Copilot Studio agent available across multiple interaction channels such as web chat, mobile applications, and Microsoft Teams while maintaining consistent functionality and security.

Generative Responses: AI-generated answers created by large language models based on user queries, knowledge sources, and system instructions rather than predefined message templates.

Pre-Deployment Controls: Security, compliance, and governance configurations that must be implemented before an agent is published and made accessible to users.

Environment-Level Enforcement: Policies and controls applied at the Power Platform environment level that govern all resources within that environment including agents, flows, and connectors.

A healthcare organization deploys a Copilot Studio agent to help patients schedule appointments, access lab results, and get general health information. Before deploying to their patient portal website and mobile app, the compliance team configures prompt modifications that include the disclaimer: "This information is for general purposes only and does not constitute medical advice. Consult your healthcare provider for specific medical guidance."

The IT security team simultaneously configures DLP policies that allow the agent to access the approved Electronic Health Record system connector and the organization's approved knowledge base, but block all other connectors including consumer cloud storage, social media platforms, and unapproved third-party APIs. This ensures patient health information never leaves the approved data ecosystem.

When a patient asks about medication side effects, the agent provides general information from the knowledge base and automatically appends the medical disclaimer. If a patient asks the agent to diagnose a condition, the prompt modifications trigger a refusal response directing them to consult a clinician. The DLP policies prevent any attempt to route conversation data to unauthorized systems, even if a future agent update tries to add new integrations.

Both controls are verified as active before the agent is published to any channel, ensuring compliance from the first patient interaction.

• Configuring compliance controls after publishing the agent, creating a window of non-compliant operation
• Using manual per-topic edits for disclaimers instead of system-level prompt modifications, leading to inconsistency and maintenance burden
• Assuming channel-level settings provide adequate security without implementing environment-level DLP policies
• Enabling Responsible AI filters per channel rather than at the platform level, resulting in uneven protection
• Testing agent functionality without validating that all governance controls are active
• Overlooking the requirement for consistent disclaimer application across generative responses
• Failing to classify connectors properly in DLP policies, leaving gaps in data protection
• Deploying to one channel before completing all pre-deployment security configurations

• Multi-select questions often have two correct answers that together form a complete solution; evaluate each option independently
• Look for keywords like "before making the agent available" which indicate pre-deployment requirements
• Statements that prohibit manual per-topic work (like "manual edits to individual topics must be avoided") are strong signals toward system-level solutions
• Channel-specific configurations are rarely the right answer when platform-level controls exist
• DLP policies and prompt modifications are complementary: one governs data access, the other governs AI behavior
• Responsible AI content moderation is a platform-level toggle, not a per-channel setting
• Pre-deployment security configuration is a recurring theme in enterprise AI deployment scenarios
• When a case study emphasizes consistency across channels, system-level controls are almost always the answer