Microsoft AB-100 Dumps
Microsoft AB-100 Practice Test 12
Question 110
During a model evaluation review, a Dynamics 365 consultant must accept a change for prompt injection boundary. The issue to be addressed is: A retrieval assistant treats an external document instruction as permission to send customer data to a remote endpoint. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the municipal services department should insert a malicious retrieved passage and verify it cannot trigger outbound customer-data transfer. Operators still need the implementation to treat retrieved text as untrusted evidence and enforce tool access outside the prompt and the preventive requirement to use scoped tool permissions and adversarial retrieval tests in the development process.
Incorrect Answers
The assumption to execute any instruction appearing in a highly ranked knowledge document is the exact failure pattern the team needs to avoid. The relevant change is to treat retrieved text as untrusted evidence and enforce tool access outside the prompt; a credible check must insert a malicious retrieved passage and verify it cannot trigger outbound customer-data transfer.
The action to use scoped tool permissions and adversarial retrieval tests in the development process can contribute to prompt injection boundary as a ongoing preventive rule step; here, a retrieval assistant treats an external document instruction as permission to send customer data to a remote endpoint. The more directly responsive choice is to insert a malicious retrieved passage and verify it cannot trigger outbound customer-data transfer.
Choosing to roll back a prompt and verify stable outputs on the affected benchmark cases would direct effort toward prompt library governance, whose characteristic issue is that several business units copy prompts into private files and produce conflicting risk-sensitive answers. The relevant corrective or preventive action here remains to insert a malicious retrieved passage and verify it cannot trigger outbound customer-data transfer, with its own validation criteria.
The action to treat retrieved text as untrusted evidence and enforce tool access outside the prompt can contribute to prompt injection boundary as a implementation step; here, a retrieval assistant treats an external document instruction as permission to send customer data to a remote endpoint. The more directly responsive choice is to insert a malicious retrieved passage and verify it cannot trigger outbound customer-data transfer.
Question 111
At a regional hospital group, the privacy engineer is preparing a customer case workflow redesign. The concrete requirement is: A high-volume in-store helper needs low-latency product lookup on modest infrastructure. Which design action BEST addresses the stated need?
Explanation
Correct Answer
The proposed implementation is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics; this matches the operational requirement that a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The team can compare small and large models on the same labeled product-query evaluation set. The team should also require model selection evidence under realistic peak workloads.
Incorrect Answers
Selecting Microsoft Foundry components with code-first orchestration and managed evaluation would direct effort toward microsoft foundry custom orchestration fit, whose characteristic issue is that an architecture team needs code-controlled model routing custom evaluation and a distributed orchestration API. The relevant corrective or preventive action here remains to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics, with its own validation criteria.
The action to compare small and large models on the same labeled product-query evaluation set can contribute to small language model selection as a targeted acceptance check step; here, a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The more directly responsive choice is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics.
The assumption to choose the largest general model without measuring whether the task needs its capacity is the exact failure pattern the team needs to avoid. The relevant change is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics; a credible check must compare small and large models on the same labeled product-query evaluation set.
Question 112
After an AI adoption planning at an insurance provider, the team observes this failure: A high-volume in-store helper needs low-latency product lookup on modest infrastructure. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?
Explanation
Correct Answer
This design mistake is to choose the largest general model without measuring whether the task needs its capacity; it conflicts with the actual small language model selection behavior needed when a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The insurance provider should instead benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics, followed by a check to compare small and large models on the same labeled product-query evaluation set.
Incorrect Answers
The alternative to compare small and large models on the same labeled product-query evaluation set is a targeted acceptance check step rather than the faulty assumption being diagnosed. The failure in small language model selection arises from choose the largest general model without measuring whether the task needs its capacity; the concrete symptom is that a high-volume in-store helper needs low-latency product lookup on modest infrastructure.
The option to allow either agent to alter the other's records through unrestricted credentials would normally serve a different cross-application architecture boundaries requirement, where a case-resolution process spans Dynamics 365 Sales and Customer Service but teams have separate data ownership. The alternative may be useful in its own context, but it does not directly supply choose the largest general model without measuring whether the task needs its capacity for the stated decision.
The alternative to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics is a implementation step rather than the faulty assumption being diagnosed. The failure in small language model selection arises from choose the largest general model without measuring whether the task needs its capacity; the concrete symptom is that a high-volume in-store helper needs low-latency product lookup on modest infrastructure.
Question 113
The software publisher is responding to a data access design. A product delivery lead discovers this issue: A high-volume in-store helper needs low-latency product lookup on modest infrastructure. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The direct repair is to route simple lookups to a validated smaller model while retaining escalation for hard cases. Once the condition is corrected, the software publisher should compare small and large models on the same labeled product-query evaluation set. To reduce recurrence, it should also require model selection evidence under realistic peak workloads; those steps provide different recovery and prevention functions.
Incorrect Answers
This approach relates to copilot studio low-code fit: a customer-service department needs an agent with business connectors and configurable conversation topics but minimal custom infrastructure. If the team chose it instead, the targeted check to compare small and large models on the same labeled product-query evaluation set could still fail because the underlying required action is to route simple lookups to a validated smaller model while retaining escalation for hard cases.
The assumption to choose the largest general model without measuring whether the task needs its capacity is the exact failure pattern the team needs to avoid. The relevant change is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics; a credible check must compare small and large models on the same labeled product-query evaluation set.
The action to compare small and large models on the same labeled product-query evaluation set can contribute to small language model selection as a targeted acceptance check step; here, a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The more directly responsive choice is to route simple lookups to a validated smaller model while retaining escalation for hard cases.
Question 114
A business application owner at a government services agency wants the following issue to be prevented in later deployments: A high-volume in-store helper needs low-latency product lookup on modest infrastructure. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
Standardizing the boundary to require model selection evidence under realistic peak workloads is preventive because the same small language model selection failure could otherwise return: a high-volume in-store helper needs low-latency product lookup on modest infrastructure. For this government services agency, the service should also benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics, and tests should compare small and large models on the same labeled product-query evaluation set so that the ongoing requirement is measurable.
Incorrect Answers
The assumption to choose the largest general model without measuring whether the task needs its capacity is the exact failure pattern the team needs to avoid. The relevant change is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics; a credible check must compare small and large models on the same labeled product-query evaluation set.
The action to establish periodic fairness review with data governance and appeal routes addresses responsible ai strategy rather than the decisive small language model selection condition. The government services agency should not infer that fixing one adjacent business-system dependency validates its unrelated workflow; the requested action is to require model selection evidence under realistic peak workloads.
The action to route simple lookups to a validated smaller model while retaining escalation for hard cases can contribute to small language model selection as a corrective remediation step; here, a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The more directly responsive choice is to require model selection evidence under realistic peak workloads.
Question 115
During an AI adoption planning, an AI governance lead must accept a change for small language model selection. The issue to be addressed is: A high-volume in-store helper needs low-latency product lookup on modest infrastructure. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
The targeted acceptance test is to compare small and large models on the same labeled product-query evaluation set. The design should first benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics. A robust manufacturing consortium deployment also needs to require model selection evidence under realistic peak workloads; passing this check provides relevant evidence but does not certify every possible input or security risk.
Incorrect Answers
For copilot studio low-code fit, it can be sensible to configure a sample customer escalation and verify the connector acts within role scope if a customer-service department needs an agent with business connectors and configurable conversation topics but minimal custom infrastructure. Without the actual change to compare small and large models on the same labeled product-query evaluation set, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
The assumption to choose the largest general model without measuring whether the task needs its capacity is the exact failure pattern the team needs to avoid. The relevant change is to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics; a credible check must compare small and large models on the same labeled product-query evaluation set.
The action to benchmark a smaller fit-for-purpose model on representative accuracy latency and cost metrics can contribute to small language model selection as a implementation step; here, a high-volume in-store helper needs low-latency product lookup on modest infrastructure. The more directly responsive choice is to compare small and large models on the same labeled product-query evaluation set.
Question 116
At a municipal services department, the AI program manager is preparing a customer service escalation review. The concrete requirement is: A manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. Which design action BEST addresses the stated need?
Explanation
Correct Answer
Choosing to validate data licensing privacy labels and evaluation feasibility before model training directly targets the described custom model feasibility gap: a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. As a lasting boundary, they also need to require data lineage approved use rights and independent holdout evaluation, because an implementation by itself does not prove safe operation.
Incorrect Answers
Choosing to adopt a documented responsible AI process with representative fairness checks and accountable remediation would direct effort toward responsible ai strategy, whose characteristic issue is that a scoring assistant may treat customers differently without evidence that disparate outcomes were tested. The relevant corrective or preventive action here remains to validate data licensing privacy labels and evaluation feasibility before model training, with its own validation criteria.
The action to inspect the training dataset provenance and reproduce metrics on unseen cases can contribute to custom model feasibility as a targeted acceptance check step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to validate data licensing privacy labels and evaluation feasibility before model training.
The assumption to start fine-tuning on uncontrolled employee messages and infer success from training loss is the exact failure pattern the team needs to avoid. The relevant change is to validate data licensing privacy labels and evaluation feasibility before model training; a credible check must inspect the training dataset provenance and reproduce metrics on unseen cases.
Question 117
After an agent reliability assessment at a global distributor, the team observes this failure: A manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?
Explanation
Correct Answer
The faulty assumption is to start fine-tuning on uncontrolled employee messages and infer success from training loss. A functioning alternative would be to validate data licensing privacy labels and evaluation feasibility before model training, and the relevant validation would be to inspect the training dataset provenance and reproduce metrics on unseen cases.
Incorrect Answers
The alternative to validate data licensing privacy labels and evaluation feasibility before model training is a implementation step rather than the faulty assumption being diagnosed. The failure in custom model feasibility arises from start fine-tuning on uncontrolled employee messages and infer success from training loss; the concrete symptom is that a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline.
The option to execute any instruction appearing in a highly ranked knowledge document would normally serve a different prompt injection boundary requirement, where a retrieval assistant treats an external document instruction as permission to send customer data to a remote endpoint. The alternative may be useful in its own context, but it does not directly supply start fine-tuning on uncontrolled employee messages and infer success from training loss for the stated decision.
The alternative to inspect the training dataset provenance and reproduce metrics on unseen cases is a targeted acceptance check step rather than the faulty assumption being diagnosed. The failure in custom model feasibility arises from start fine-tuning on uncontrolled employee messages and infer success from training loss; the concrete symptom is that a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline.
Question 118
The digital payments provider is responding to a preproduction release gate. An integration engineer discovers this issue: A manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The appropriate remediation is to remove unauthorized data and restart the proof of concept with governed labeled samples, because the failure concerns custom model feasibility: a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. A meaningful verification should inspect the training dataset provenance and reproduce metrics on unseen cases, and the durable operating rule should require data lineage approved use rights and independent holdout evaluation after the immediate fix.
Incorrect Answers
The action to inspect the training dataset provenance and reproduce metrics on unseen cases can contribute to custom model feasibility as a targeted acceptance check step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to remove unauthorized data and restart the proof of concept with governed labeled samples.
This approach relates to responsible ai strategy: a scoring assistant may treat customers differently without evidence that disparate outcomes were tested. If the team chose it instead, the targeted check to inspect the training dataset provenance and reproduce metrics on unseen cases could still fail because the underlying required action is to remove unauthorized data and restart the proof of concept with governed labeled samples.
The assumption to start fine-tuning on uncontrolled employee messages and infer success from training loss is the exact failure pattern the team needs to avoid. The relevant change is to validate data licensing privacy labels and evaluation feasibility before model training; a credible check must inspect the training dataset provenance and reproduce metrics on unseen cases.
Question 119
A Copilot Studio architect at a financial services firm wants the following issue to be prevented in later deployments: A manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
The sustained prevention measure is to require data lineage approved use rights and independent holdout evaluation. The implementation can still use the separate action to validate data licensing privacy labels and evaluation feasibility before model training, but a guardrail constrains future changes. The financial services firm can measure whether the rule works by arranging to inspect the training dataset provenance and reproduce metrics on unseen cases.
Incorrect Answers
Choosing to require environment governance and clear ownership for low-code agent changes would direct effort toward copilot studio low-code fit, whose characteristic issue is that a customer-service department needs an agent with business connectors and configurable conversation topics but minimal custom infrastructure. The relevant corrective or preventive action here remains to require data lineage approved use rights and independent holdout evaluation, with its own validation criteria.
The action to inspect the training dataset provenance and reproduce metrics on unseen cases can contribute to custom model feasibility as a targeted acceptance check step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to require data lineage approved use rights and independent holdout evaluation.
The assumption to start fine-tuning on uncontrolled employee messages and infer success from training loss is the exact failure pattern the team needs to avoid. The relevant change is to validate data licensing privacy labels and evaluation feasibility before model training; a credible check must inspect the training dataset provenance and reproduce metrics on unseen cases.
The action to validate data licensing privacy labels and evaluation feasibility before model training can contribute to custom model feasibility as a implementation step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to require data lineage approved use rights and independent holdout evaluation.
The action to remove unauthorized data and restart the proof of concept with governed labeled samples can contribute to custom model feasibility as a corrective remediation step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to require data lineage approved use rights and independent holdout evaluation.
Question 120
During an agent reliability assessment, a Power Platform administrator must accept a change for custom model feasibility. The issue to be addressed is: A manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the professional services partnership should inspect the training dataset provenance and reproduce metrics on unseen cases. Operators still need the implementation to validate data licensing privacy labels and evaluation feasibility before model training and the preventive requirement to require data lineage approved use rights and independent holdout evaluation.
Incorrect Answers
For microsoft 365 copilot extension choice, it can be sensible to test two users with different SharePoint access against the same specialist query if employees need a Microsoft 365-grounded specialist assistant that follows enterprise file permissions. Without the actual change to inspect the training dataset provenance and reproduce metrics on unseen cases, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
The assumption to start fine-tuning on uncontrolled employee messages and infer success from training loss is the exact failure pattern the team needs to avoid. The relevant change is to validate data licensing privacy labels and evaluation feasibility before model training; a credible check must inspect the training dataset provenance and reproduce metrics on unseen cases.
The action to validate data licensing privacy labels and evaluation feasibility before model training can contribute to custom model feasibility as a implementation step; here, a manufacturer proposes fine-tuning a model without rights to its training data and without a labeled quality baseline. The more directly responsive choice is to inspect the training dataset provenance and reproduce metrics on unseen cases.
Agentic AI Business Solutions Architect (AB-100) Practice Tests
- Microsoft AB-100 Practice Test 10 - 10 Questions (Last Reviewed 12-Nov-2026)
- Microsoft AB-100 Practice Test 11 - 10 Questions (Last Reviewed 11-Nov-2026)
- Microsoft AB-100 Practice Test 12 - 10 Questions (Last Reviewed 10-Nov-2026)
- Microsoft AB-100 Practice Test 13 - 10 Questions (Last Reviewed 09-Nov-2026)
- Microsoft AB-100 Practice Test 14 - 10 Questions (Last Reviewed 08-Nov-2026)