A is incorrect: Allowing the model to call any external website with full customer details violates the requirement of protecting personal data in prompts and outputs. This approach also raises privacy concerns and does not align with responsible AI standards.

B is incorrect: Disabling all logging to prevent storing personal data may seem like a way to protect privacy, but it contradicts the requirement of providing a way to trace and review AI-assisted responses when an incident occurs. Logging is essential for accountability and transparency in AI interactions.

C is incorrect: Sharing conversation logs publicly compromises customer privacy and violates the requirement of preventing the model from returning unsafe or disallowed content to customers. This approach also undermines the need for access controls and confidentiality in AI systems.

D is correct: Combining content filters, data minimization, access controls, and robust auditing aligns with responsible AI standards by addressing all the specified requirements. This approach ensures that the model only provides safe content, protects personal data, restricts access to authorized staff, and enables traceability and review of AI-assisted responses.

A is correct: Yes, the mapping of requirements to the Researcher and Analyst agents aligns with how these agents are intended to be used. The Researcher agent is designed for producing structured, citation-rich briefings that blend clinical guidelines, policy docs, and research articles, which aligns with Requirement R1. The Analyst agent is intended for analyzing datasets, finding trends, outliers, and key drivers, which aligns with Requirement R2. The proposed solution effectively matches the agents with the specific tasks outlined in the requirements.

A is incorrect: Replacing Microsoft 365 Copilot with a custom Azure OpenAI chatbot hosted on a separate website would not align with the Microsoft 365 Copilot extensibility framework, as it goes against the requirement of keeping the experience inside Microsoft 365 Copilot and Teams.

B is incorrect: Only improving prompts in the base Copilot experience and avoiding any integration work does not align with the requirements of answering questions using data from a third-party CRM and ticketing system, and triggering multi-step workflows from Copilot chats. This approach does not extend the capabilities of Microsoft 365 Copilot as needed.

C is incorrect: Creating a new Power BI workspace and sharing CRM dashboards there does not align with the requirement of extending Microsoft 365 Copilot to bring in external data and actions. Power BI is not the appropriate tool for integrating third-party CRM and ticketing system data into Copilot chats.

D is correct: Extending Microsoft 365 Copilot using Copilot connectors and agents via Copilot Studio and related tools aligns with the Microsoft 365 Copilot extensibility framework. This approach allows for integrating external data sources and triggering actions from Copilot chats while keeping the experience inside Microsoft 365 Copilot and Teams, as required by Fabrikam.

A is incorrect: Building a custom AI app on Azure OpenAI would require a significant engineering project and may not show clear value in the short term. Rolling it out separately from Microsoft 365 may also create integration challenges and hinder user adoption.

B is correct: Rolling out Microsoft 365 Copilot out of the box first allows Contoso to quickly leverage AI capabilities without the need for a large engineering project. Planning extensibility in a later phase ensures that the solution can be customized to meet specific business needs.

C is incorrect: Immediately building custom agents and connectors before enabling any Copilot features may lead to unnecessary complexity and delays in realizing productivity gains. It is important to first leverage out-of-the-box solutions before investing in custom development.

D is incorrect: Blocking Copilot and waiting until all business systems are modeled as custom agents may result in missed opportunities to improve productivity in the short term. It is more effective to start with available AI features and gradually integrate custom solutions as needed.

A is incorrect: Training a custom model using Azure Machine Learning would require expertise in AI engineering and managing models, which Contoso's HR team wants to avoid. This option involves significant infrastructure and model management, making it less suitable for their specific needs.

B is correct: Using Microsoft Copilot for Microsoft 365 allows Contoso's HR team to draft job descriptions and rewrite internal policy documents directly inside Word and Outlook without the need for managing models, prompts, or infrastructure. It provides a user-friendly and efficient solution for their requirements.

C is incorrect: Deploying Azure OpenAI models behind a custom web app involves more complexity and infrastructure management than Contoso's HR team is looking for. This option would require additional development resources and expertise, which goes against their goal of avoiding model and infrastructure management.

D is incorrect: Configuring Azure AI Search with semantic ranking only focuses on improving search results based on the meaning of the content. While it may enhance search functionality, it does not directly address Contoso's HR team's need for generating job descriptions and rewriting policy documents within Word and Outlook without managing models or infrastructure.

A is incorrect: An AI champions program is not primarily focused on budget approval or vendor contracts. Its main purpose is to promote and advocate for AI within the organization, rather than handling financial decisions.

B is correct: This choice accurately describes an AI champions program as a community of enthusiastic users who promote AI, provide coaching to peers, and share real-world insights to drive adoption and success within the organization.

C is incorrect: A helpdesk team that only fixes AI-related technical incidents is more aligned with a technical support function rather than the broader scope of activities involved in an AI champions program.

D is incorrect: A committee that solely focuses on writing AI policies and does not engage with end users or promote AI adoption does not align with the collaborative and user-focused nature of an AI champions program.

A is incorrect: The speed of AI systems compared to human decision-making is not directly related to the principles of responsible AI, such as fairness, reliability, safety, privacy, inclusiveness, transparency, and accountability. While speed can be a factor in certain applications, it is not the primary focus of building trust through responsible AI.

B is incorrect: While autonomy is a characteristic of AI systems, the idea that they should be fully autonomous and require no human involvement contradicts the principles of responsible AI. Human oversight and intervention are crucial for ensuring that AI systems align with organizational values, are safe for people and society, and uphold ethical standards.

C is correct: Trustworthiness, alignment with organizational values, and safety for people and society are key outcomes of applying the principles of responsible AI. By designing, testing, and deploying AI systems with fairness, reliability, privacy, inclusiveness, transparency, and accountability in mind, organizations can build trust with stakeholders and ensure that their AI technologies benefit society.

D is incorrect: While minimizing errors is an important goal in AI system design, guaranteeing that AI systems will never make errors in any situation is unrealistic. The focus of responsible AI is on mitigating risks, ensuring safety, and upholding ethical standards, rather than eliminating errors entirely. Trust is built through transparency, accountability, and a commitment to ethical practices.

A is incorrect: GitHub Copilot is an AI-powered code completion tool designed for developers to assist in writing code. It is not specifically tailored for building and managing custom AI agents, connecting to internal data sources, or deploying to channels like web and Microsoft Teams.

B is incorrect: Microsoft Security Copilot is a security tool that helps organizations protect against cyber threats. It is not designed for building and managing custom AI agents, connecting to internal data sources, or deploying to channels like web and Microsoft Teams.

C is incorrect: Microsoft 365 Copilot is not a product offered by Microsoft. It does not exist in the Microsoft product lineup, so it cannot be the correct choice for building and managing custom AI agents, connecting to internal data sources, and deploying to channels like web and Microsoft Teams.

D is correct: Microsoft Copilot Studio is a graphical, low-code environment designed for building and managing custom AI agents, connecting to internal data sources, using natural language to automate tasks, and deploying to channels like web and Microsoft Teams. It is the best fit for the scenario described in the question.

A is correct: This statement accurately reflects the benefits of implementing responsible AI practices within organizations. By focusing on responsible AI, organizations can reduce risks such as harm and non-compliance while also improving business outcomes like user trust, system quality, and adoption. This alignment between risk reduction and positive business outcomes is a key aspect of responsible AI implementation.

A is correct: Azure AI services offer enterprise-grade security and compliance features, including encryption in transit, data residency options, RBAC, private networking, and alignment with regulatory baselines. These features ensure that customer data is protected, processed in compliance with regional rules, and access is controlled securely, meeting the bank's key requirements.

B is incorrect: Public unauthenticated HTTP endpoints that accept requests from the open internet do not align with the bank's requirements for secure data processing, compliance with regional rules, and controlled access via RBAC and private networking. This choice does not address the key needs of the global bank.

C is incorrect: Having a single shared admin account for all apps may simplify permission management, but it does not address the specific security, compliance, data residency, and access control requirements outlined by the global bank. This choice does not align with the bank's key needs for using Azure AI services.

D is incorrect: Automatically sharing prompts and completions with other tenants to improve global model quality does not directly address the bank's requirements for secure data processing, compliance with regional rules, controlled access via RBAC and private networking, and alignment with Azure security baselines and regulatory compliance. This choice does not meet the key needs of the global bank.