10 realistic exam questions, graded as you go, with the exact rule or statute citation for every answer. Start free, no sign-up.
Timed mock exam
40 questions in 48 minutes, drawn from every topic. No answers shown until the end, then a full score breakdown, your weakest topics and a review of what you missed.
Prices are in US dollars, plus tax where it applies. Stripe, our payment processor, calculates tax at checkout.
All sales are final. If something is really wrong, such as a duplicate charge or a purchase that does not unlock what you bought, write to support@careerexampro.com and we will put it right.
One-time payment. Nothing renews.
You have an exam ticket for this practice test. Start the timed mock exam to use it.
An Azure AI Foundry customer-service agent summarizes incoming vendor emails and extracts invoice fields. A malicious vendor hides white-on-white instructions inside an email body telling the agent to forward internal customer records to an external address. Which Azure AI Content Safety control is specifically designed to detect and block this attack vector?
Prompt Shields for indirect attacks (XPIA), scanning untrusted documents and emails before model execution.
Azure DDoS Protection Standard, dropping malformed TCP SYN floods at the enterprise email relay before delivery.
Protected Material Detection for code, which compares generated Python snippets against public GitHub repositories for open-source license matches.
Azure AI Translator glossaries stripping non-English unicode punctuation from invoices before PDF rendering.
Show the answer
A. Prompt Shields for indirect attacks (XPIA), scanning untrusted documents and emails before model execution.
Azure AI Content Safety Prompt Shields distinguish between direct user-prompt attacks (jailbreaks) and indirect prompt injection attacks (XPIA) embedded inside third-party data such as emails, web pages, or RAG documents. Enabling indirect-attack Prompt Shields flags and blocks adversarial instructions hidden in untrusted grounding content.
Reference: Microsoft AI-103 Study Guide / Azure AI Content Safety Prompt Shields Documentation
A financial research agent currently uses a single-model ReAct loop on a large frontier LLM, causing high latency and cost because it re-reasons over the entire growing conversation history after every simple API lookup. How does migrating to a Plan-and-Execute architecture address this bottleneck?
A planner creates a multi-step plan upfront, smaller workers execute steps, and replanning triggers only on exceptions.
It removes all tool-calling interfaces and pre-generates every analytical answer into a static cache before queries arrive.
It forces the frontier model to emit all intermediate observations inside hidden reasoning tokens without calling APIs.
It replaces structured reasoning with unguided zero-shot completion so every financial calculation completes in a single forward pass with fixed latency.
Show the answer
A. A planner creates a multi-step plan upfront, smaller workers execute steps, and replanning triggers only on exceptions.
Plan-and-Execute decouples high-level planning from step execution: a capable planner creates a structured step list upfront, lightweight/cheaper models execute independent steps (often concurrently), and the planner is re-invoked only when results require replanning. This sharply reduces redundant frontier-model calls and latency compared to step-by-step ReAct loops.
A retail organization wants two distinct Microsoft 365 agents: (1) a quick-launch IT Helpdesk assistant embedded inside Microsoft 365 Copilot that grounds answers in SharePoint policies and calls ServiceNow APIs using Microsoft's built-in orchestrator; and (2) an external-facing supply-chain agent that must run a custom fine-tuned open-weight model with bespoke graph orchestration. Which Copilot Studio / Azure pattern fits each requirement?
Build both as custom-engine agents on bare-metal GPUs, since declarative agents cannot read SharePoint or call APIs.
Build both as declarative agents inside Microsoft 365 Copilot, because declarative agents allow swapping out the base model for any custom fine-tuned weights.
Build the IT Helpdesk assistant as a declarative agent extending M365 Copilot, and build the supply-chain assistant as a custom-engine agent.
Build the IT Helpdesk assistant as a custom-engine agent in Azure Kubernetes Service, and deploy the supply-chain model as a declarative M365 plugin.
Show the answer
C. Build the IT Helpdesk assistant as a declarative agent extending M365 Copilot, and build the supply-chain assistant as a custom-engine agent.
Declarative agents extend Microsoft 365 Copilot's built-in orchestrator and foundation model with scoped instructions, SharePoint/Graph grounding, and API actions—ideal for fast, governed M365 workflows. Custom-engine agents let developers supply their own LLM/SLM, bespoke orchestration logic, and custom RAG pipelines across standalone or Teams channels.
Reference: Microsoft Copilot Studio & Microsoft 365 Agents Extensibility Guide (AI-103 / AB-730)
A multi-session customer operations agent needs to remember three distinct categories of state: (1) messages within the active troubleshooting thread, (2) confirmed account preferences and product entitlements, and (3) prior resolution trajectories that worked for similar outage tickets. How should these map to agent memory layers?
Store active messages in semantic memory, account preferences in GPU KV-cache memory, and prior trajectories solely in the model's fine-tuned weights.
Active messages in short-term working memory, account facts in semantic memory, and past trajectories in episodic memory.
Concatenate all three categories inside one static system prompt shared globally across every tenant session.
Store active messages in episodic memory, product entitlements in short-term scratchpad memory, and past trajectories in volatile browser cookies.
Show the answer
B. Active messages in short-term working memory, account facts in semantic memory, and past trajectories in episodic memory.
Agentic architectures separate short-term working memory (current thread state/checkpoint) from long-term memory, which subdivides into semantic memory (distilled facts, user preferences, entity graphs) and episodic memory (indexed past interaction trajectories and outcomes retrieved by task similarity to guide future decisions).
A regional sales operations team wants to standardize weekly account-renewal briefs using Microsoft 365 Copilot (AB-730). Several reps have been pasting customer contracts into free consumer web chatbots on personal accounts. Which business-user practice ensures repeatable quality while maintaining enterprise data protection?
Use Copilot with Commercial Data Protection, structured prompt templates grounded in approved files, and human review.
Permit free consumer chatbots provided reps manually delete the browser history tab immediately after copying the generated renewal brief into PowerPoint.
Rely on brief ad-hoc prompts in M365 Copilot and email generated renewal pricing directly to clients without review.
Export all customer contracts into a single public anonymous link so any third-party AI summarization plugin can parse them without Entra ID sign-in.
Show the answer
A. Use Copilot with Commercial Data Protection, structured prompt templates grounded in approved files, and human review.
Enterprise AI productivity (AB-730) requires keeping confidential data inside the organization's Commercial Data Protection / M365 tenant boundary (never pasting client contracts into unmanaged consumer AI tools), standardizing outputs via structured reusable prompt templates grounded in authoritative files, and retaining human-in-the-loop review before external customer communication.
Reference: Microsoft AB-730 (AI Business Professional) Study Guide / Enterprise Data Protection & Prompt Design