Practice Exams:

Microsoft AB-100 Dumps

EXAM: Agentic AI Business Solutions Architect
Verified By Experts
Premium Bundle
$49.99

AB-100 Premium ETE File


  • Premium File 129 Questions & Answers
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Microsoft AB-100 Practice Test 63

Question 620

An enterprise solution architect at a university network is conducting a preproduction release gate concerning agent least-privilege tool scope. The observed problem is: A service agent can read cases but also delete customer accounts through its broadly privileged tool. Which TWO actions together represent the needed implementation AND a durable preventive safeguard? (Select TWO answers.)

  1. Enforce service-level authorization independent of conversation instructions
  2. Manage distinct workload identities with expiry review and revocation when agents retire
  3. Give the agent owner privileges because it has to handle complex queries
  4. Grant only the tool actions and record scope required for the business purpose
  5. Try to delete a customer as a read-only service agent and verify denial

Explanation

Correct Answers

A:

Standardizing the boundary to enforce service-level authorization independent of conversation instructions is preventive because the same agent least-privilege tool scope failure could otherwise return: a service agent can read cases but also delete customer accounts through its broadly privileged tool. For this university network, the service should also grant only the tool actions and record scope required for the business purpose, and tests should try to delete a customer as a read-only service agent and verify denial so that the ongoing requirement is measurable.

D:

Choosing to grant only the tool actions and record scope required for the business purpose directly targets the described agent least-privilege tool scope gap: a service agent can read cases but also delete customer accounts through its broadly privileged tool. As a lasting boundary, they also need to enforce service-level authorization independent of conversation instructions, because an implementation by itself does not prove safe operation.

Incorrect Answers

B:

Choosing to manage distinct workload identities with expiry review and revocation when agents retire would direct effort toward agent identity lifecycle, whose characteristic issue is that an unattended service principal continues using broad permissions after an agent project is discontinued. The relevant corrective or preventive action here remains to grant only the tool actions and record scope required for the business purpose, with its own validation criteria.

C:

The assumption to give the agent owner privileges because it has to handle complex queries is the exact failure pattern the team needs to avoid. The relevant change is to grant only the tool actions and record scope required for the business purpose; a credible check must try to delete a customer as a read-only service agent and verify denial.

E:

The action to try to delete a customer as a read-only service agent and verify denial can contribute to agent least-privilege tool scope as a targeted acceptance check step; here, a service agent can read cases but also delete customer accounts through its broadly privileged tool. The more directly responsive choice is to grant only the tool actions and record scope required for the business purpose.

Question 621

At an education technology firm, the AI governance lead is preparing a business process modernization. The concrete requirement is: An attacker puts forged approval instructions inside a retrieved supplier invoice. Which design action BEST addresses the stated need?

  1. Inject a forged approval in an invoice and confirm payment authority is not granted
  2. Obey an invoice paragraph that claims to override the agent's tool access policies
  3. Treat document text as untrusted data and guard external actions with independent policy checks
  4. Sanitize approve and restrict tuning data with lineage and appropriate rights

Explanation

Correct Answer

C:

The proposed implementation is to treat document text as untrusted data and guard external actions with independent policy checks; this matches the operational requirement that an attacker puts forged approval instructions inside a retrieved supplier invoice. The team can inject a forged approval in an invoice and confirm payment authority is not granted. The team should also run adversarial retrieved-content tests and minimize tool privilege.

Incorrect Answers

A:

The action to inject a forged approval in an invoice and confirm payment authority is not granted can contribute to prompt manipulation defense as a targeted acceptance check step; here, an attacker puts forged approval instructions inside a retrieved supplier invoice. The more directly responsive choice is to treat document text as untrusted data and guard external actions with independent policy checks.

B:

The assumption to obey an invoice paragraph that claims to override the agent's tool access policies is the exact failure pattern the team needs to avoid. The relevant change is to treat document text as untrusted data and guard external actions with independent policy checks; a credible check must inject a forged approval in an invoice and confirm payment authority is not granted.

D:

Choosing to sanitize approve and restrict tuning data with lineage and appropriate rights would direct effort toward sensitive model tuning data, whose characteristic issue is that a customization pipeline accepts raw customer complaint text including credentials as tuning examples. The relevant corrective or preventive action here remains to treat document text as untrusted data and guard external actions with independent policy checks, with its own validation criteria.

Question 622

After a security readiness checkpoint at a customer support outsourcer, the team observes this failure: An attacker puts forged approval instructions inside a retrieved supplier invoice. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Inject a forged approval in an invoice and confirm payment authority is not granted
  2. Treat delegation as unconditional permission escalation for downstream agents
  3. Treat document text as untrusted data and guard external actions with independent policy checks
  4. Obey an invoice paragraph that claims to override the agent's tool access policies

Explanation

Correct Answer

D:

This design mistake is to obey an invoice paragraph that claims to override the agent's tool access policies; it conflicts with the actual prompt manipulation defense behavior needed when an attacker puts forged approval instructions inside a retrieved supplier invoice. The customer support outsourcer should instead treat document text as untrusted data and guard external actions with independent policy checks, followed by a check to inject a forged approval in an invoice and confirm payment authority is not granted.

Incorrect Answers

A:

The alternative to inject a forged approval in an invoice and confirm payment authority is not granted is a targeted acceptance check step rather than the faulty assumption being diagnosed. The failure in prompt manipulation defense arises from obey an invoice paragraph that claims to override the agent's tool access policies; the concrete symptom is that an attacker puts forged approval instructions inside a retrieved supplier invoice.

B:

The option to treat delegation as unconditional permission escalation for downstream agents would normally serve a different multi-agent delegated authority requirement, where a planning agent delegates tasks to an executor agent with more privileges than the original user. The alternative may be useful in its own context, but it does not directly supply obey an invoice paragraph that claims to override the agent's tool access policies for the stated decision.

C:

The alternative to treat document text as untrusted data and guard external actions with independent policy checks is a implementation step rather than the faulty assumption being diagnosed. The failure in prompt manipulation defense arises from obey an invoice paragraph that claims to override the agent's tool access policies; the concrete symptom is that an attacker puts forged approval instructions inside a retrieved supplier invoice.

Question 623

The national charity is responding to an incident response exercise. A contact center architect discovers this issue: An attacker puts forged approval instructions inside a retrieved supplier invoice. What corrective step should the team take to restore the intended functionality most directly?

  1. Repair inaccessible interaction paths and re-run representative user tests
  2. Quarantine the malicious source and revoke any unauthorized action
  3. Inject a forged approval in an invoice and confirm payment authority is not granted
  4. Obey an invoice paragraph that claims to override the agent's tool access policies

Explanation

Correct Answer

B:

The direct repair is to quarantine the malicious source and revoke any unauthorized action. Once the condition is corrected, the national charity should inject a forged approval in an invoice and confirm payment authority is not granted. To reduce recurrence, it should also run adversarial retrieved-content tests and minimize tool privilege; those steps provide different recovery and prevention functions.

Incorrect Answers

A:

This approach relates to responsible ai review of accessibility: an AI-powered customer portal works for typed English input but fails users relying on accessibility features. If the team chose it instead, the targeted check to inject a forged approval in an invoice and confirm payment authority is not granted could still fail because the underlying required action is to quarantine the malicious source and revoke any unauthorized action.

C:

The action to inject a forged approval in an invoice and confirm payment authority is not granted can contribute to prompt manipulation defense as a targeted acceptance check step; here, an attacker puts forged approval instructions inside a retrieved supplier invoice. The more directly responsive choice is to quarantine the malicious source and revoke any unauthorized action.

D:

The assumption to obey an invoice paragraph that claims to override the agent's tool access policies is the exact failure pattern the team needs to avoid. The relevant change is to treat document text as untrusted data and guard external actions with independent policy checks; a credible check must inject a forged approval in an invoice and confirm payment authority is not granted.

Question 624

A product delivery lead at a housing association wants the following issue to be prevented in later deployments: An attacker puts forged approval instructions inside a retrieved supplier invoice. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Quarantine the malicious source and revoke any unauthorized action
  2. Obey an invoice paragraph that claims to override the agent's tool access policies
  3. Run adversarial retrieved-content tests and minimize tool privilege
  4. Test regulated output classes and monitor leakage incidents without storing extra secrets

Explanation

Correct Answer

C:

Standardizing the boundary to run adversarial retrieved-content tests and minimize tool privilege is preventive because the same prompt manipulation defense failure could otherwise return: an attacker puts forged approval instructions inside a retrieved supplier invoice. For this housing association, the service should also treat document text as untrusted data and guard external actions with independent policy checks, and tests should inject a forged approval in an invoice and confirm payment authority is not granted so that the ongoing requirement is measurable.

Incorrect Answers

A:

The action to quarantine the malicious source and revoke any unauthorized action can contribute to prompt manipulation defense as a corrective remediation step; here, an attacker puts forged approval instructions inside a retrieved supplier invoice. The more directly responsive choice is to run adversarial retrieved-content tests and minimize tool privilege.

B:

The assumption to obey an invoice paragraph that claims to override the agent's tool access policies is the exact failure pattern the team needs to avoid. The relevant change is to treat document text as untrusted data and guard external actions with independent policy checks; a credible check must inject a forged approval in an invoice and confirm payment authority is not granted.

D:

The action to test regulated output classes and monitor leakage incidents without storing extra secrets addresses model output data leakage rather than the decisive prompt manipulation defense condition. The housing association should not infer that fixing one adjacent business-system dependency validates its unrelated workflow; the requested action is to run adversarial retrieved-content tests and minimize tool privilege.

Question 625

During a security readiness checkpoint, a business application owner must accept a change for prompt manipulation defense. The issue to be addressed is: An attacker puts forged approval instructions inside a retrieved supplier invoice. Which concrete verification action demonstrates that the relevant behavior works?

  1. Challenge a rejection and verify the record links to actual criteria and evidence
  2. Inject a forged approval in an invoice and confirm payment authority is not granted
  3. Obey an invoice paragraph that claims to override the agent's tool access policies
  4. Treat document text as untrusted data and guard external actions with independent policy checks

Explanation

Correct Answer

B:

The targeted acceptance test is to inject a forged approval in an invoice and confirm payment authority is not granted. The design should first treat document text as untrusted data and guard external actions with independent policy checks. A robust retail franchise deployment also needs to run adversarial retrieved-content tests and minimize tool privilege; passing this check provides relevant evidence but does not certify every possible input or security risk.

Incorrect Answers

A:

For ai accountability and explanation, it can be sensible to challenge a rejection and verify the record links to actual criteria and evidence if an automated supplier rejection cannot identify the business policy or evidence that caused the result. Without the actual change to inject a forged approval in an invoice and confirm payment authority is not granted, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

C:

The assumption to obey an invoice paragraph that claims to override the agent's tool access policies is the exact failure pattern the team needs to avoid. The relevant change is to treat document text as untrusted data and guard external actions with independent policy checks; a credible check must inject a forged approval in an invoice and confirm payment authority is not granted.

D:

The action to treat document text as untrusted data and guard external actions with independent policy checks can contribute to prompt manipulation defense as a implementation step; here, an attacker puts forged approval instructions inside a retrieved supplier invoice. The more directly responsive choice is to inject a forged approval in an invoice and confirm payment authority is not granted.

Question 626

At a university network, the Power Platform administrator is preparing a cross-platform change window. The concrete requirement is: A customer-support agent can send unrestricted HTTP requests containing private case content. Which design action BEST addresses the stated need?

  1. Restrict outbound destinations payload fields and authorization of network tools
  2. Allow any user-provided URL to receive the full customer record for debugging
  3. Document authorized time-bounded exceptions with explicit owner and mitigation evidence
  4. Ask the agent to post a private case to a hostile URL and confirm denial

Explanation

Correct Answer

A:

Choosing to restrict outbound destinations payload fields and authorization of network tools directly targets the described data exfiltration through tools gap: a customer-support agent can send unrestricted HTTP requests containing private case content. As a lasting boundary, they also need to use destination allowlists DLP and redaction at connector boundaries, because an implementation by itself does not prove safe operation.

Incorrect Answers

B:

The assumption to allow any user-provided URL to receive the full customer record for debugging is the exact failure pattern the team needs to avoid. The relevant change is to restrict outbound destinations payload fields and authorization of network tools; a credible check must ask the agent to post a private case to a hostile URL and confirm denial.

C:

Choosing to document authorized time-bounded exceptions with explicit owner and mitigation evidence would direct effort toward agent risk exception governance, whose characteristic issue is that a business owner accepts residual risk for an agent without specifying duration or compensating measures. The relevant corrective or preventive action here remains to restrict outbound destinations payload fields and authorization of network tools, with its own validation criteria.

D:

The action to ask the agent to post a private case to a hostile URL and confirm denial can contribute to data exfiltration through tools as a targeted acceptance check step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to restrict outbound destinations payload fields and authorization of network tools.

Question 627

After a business AI architecture review at a logistics enterprise, the team observes this failure: A customer-support agent can send unrestricted HTTP requests containing private case content. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Use destination allowlists DLP and redaction at connector boundaries
  2. Ask the agent to post a private case to a hostile URL and confirm denial
  3. Restrict outbound destinations payload fields and authorization of network tools
  4. Measure only average answer quality while excluding affected customer groups
  5. Allow any user-provided URL to receive the full customer record for debugging

Explanation

Correct Answer

E:

The faulty assumption is to allow any user-provided URL to receive the full customer record for debugging. A functioning alternative would be to restrict outbound destinations payload fields and authorization of network tools, and the relevant validation would be to ask the agent to post a private case to a hostile URL and confirm denial.

Incorrect Answers

A:

The alternative to use destination allowlists DLP and redaction at connector boundaries is a ongoing preventive rule step rather than the faulty assumption being diagnosed. For the root-cause diagnosis decision on data exfiltration through tools with the option “Use destination allowlists DLP and redaction at connector boundaries”, the failure in data exfiltration through tools arises from allow any user-provided URL to receive the full customer record for debugging; the concrete symptom is that a customer-support agent can send unrestricted HTTP requests containing private case content.

B:

The alternative to ask the agent to post a private case to a hostile URL and confirm denial is a targeted acceptance check step rather than the faulty assumption being diagnosed. For the root-cause diagnosis decision on data exfiltration through tools with the option “Ask the agent to post a private case to a hostile URL and confirm denial”, the failure in data exfiltration through tools arises from allow any user-provided URL to receive the full customer record for debugging; the concrete symptom is that a customer-support agent can send unrestricted HTTP requests containing private case content.

C:

The alternative to restrict outbound destinations payload fields and authorization of network tools is a implementation step rather than the faulty assumption being diagnosed. For the root-cause diagnosis decision on data exfiltration through tools with the option “Restrict outbound destinations payload fields and authorization of network tools”, the failure in data exfiltration through tools arises from allow any user-provided URL to receive the full customer record for debugging; the concrete symptom is that a customer-support agent can send unrestricted HTTP requests containing private case content.

D:

For the root-cause diagnosis decision on data exfiltration through tools with the option “Measure only average answer quality while excluding affected customer groups”, this approach relates to responsible ai harm assessment: a customer-help agent may give discriminatory finance guidance to vulnerable customers. If the team chose it instead, the targeted check to ask the agent to post a private case to a hostile URL and confirm denial could still fail because the underlying required action is to allow any user-provided URL to receive the full customer record for debugging.

Question 628

The pharmaceutical supplier is responding to a solution integration review. A business systems designer discovers this issue: A customer-support agent can send unrestricted HTTP requests containing private case content. What corrective step should the team take to restore the intended functionality most directly?

  1. Disable the unsafe tool and review traces for leaked case details
  2. Quarantine the unauthorized examples and rebuild the dataset with controlled provenance
  3. Allow any user-provided URL to receive the full customer record for debugging
  4. Ask the agent to post a private case to a hostile URL and confirm denial

Explanation

Correct Answer

A:

The appropriate remediation is to disable the unsafe tool and review traces for leaked case details, because the failure concerns data exfiltration through tools: a customer-support agent can send unrestricted HTTP requests containing private case content. A meaningful verification should ask the agent to post a private case to a hostile URL and confirm denial, and the durable operating rule should use destination allowlists DLP and redaction at connector boundaries after the immediate fix.

Incorrect Answers

B:

This approach relates to sensitive model tuning data: a customization pipeline accepts raw customer complaint text including credentials as tuning examples. If the team chose it instead, the targeted check to ask the agent to post a private case to a hostile URL and confirm denial could still fail because the underlying required action is to disable the unsafe tool and review traces for leaked case details.

C:

The assumption to allow any user-provided URL to receive the full customer record for debugging is the exact failure pattern the team needs to avoid. The relevant change is to restrict outbound destinations payload fields and authorization of network tools; a credible check must ask the agent to post a private case to a hostile URL and confirm denial.

D:

The action to ask the agent to post a private case to a hostile URL and confirm denial can contribute to data exfiltration through tools as a targeted acceptance check step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to disable the unsafe tool and review traces for leaked case details.

Question 629

An integration engineer at a regional hospital group wants the following issue to be prevented in later deployments: A customer-support agent can send unrestricted HTTP requests containing private case content. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Ask the agent to post a private case to a hostile URL and confirm denial
  2. Use destination allowlists DLP and redaction at connector boundaries
  3. Disable the unsafe tool and review traces for leaked case details
  4. Restrict outbound destinations payload fields and authorization of network tools
  5. Set data-classified audit schemas and retention schedules before production
  6. Allow any user-provided URL to receive the full customer record for debugging

Explanation

Correct Answer

B:

The sustained prevention measure is to use destination allowlists DLP and redaction at connector boundaries. The implementation can still use the separate action to restrict outbound destinations payload fields and authorization of network tools, but a guardrail constrains future changes. The regional hospital group can measure whether the rule works by arranging to ask the agent to post a private case to a hostile URL and confirm denial.

Incorrect Answers

A:

The action to ask the agent to post a private case to a hostile URL and confirm denial can contribute to data exfiltration through tools as a targeted acceptance check step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to use destination allowlists DLP and redaction at connector boundaries.

C:

The action to disable the unsafe tool and review traces for leaked case details can contribute to data exfiltration through tools as a corrective remediation step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to use destination allowlists DLP and redaction at connector boundaries.

D:

The action to restrict outbound destinations payload fields and authorization of network tools can contribute to data exfiltration through tools as a implementation step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to use destination allowlists DLP and redaction at connector boundaries.

E:

Choosing to set data-classified audit schemas and retention schedules before production would direct effort toward retention minimization for agent logs, whose characteristic issue is that aI tool traces retain full chat histories indefinitely even though incidents need only minimal evidence. The relevant corrective or preventive action here remains to use destination allowlists DLP and redaction at connector boundaries, with its own validation criteria.

F:

The assumption to allow any user-provided URL to receive the full customer record for debugging is the exact failure pattern the team needs to avoid. The relevant change is to restrict outbound destinations payload fields and authorization of network tools; a credible check must ask the agent to post a private case to a hostile URL and confirm denial.

Question 630

During a business AI architecture review, a Copilot Studio architect must accept a change for data exfiltration through tools. The issue to be addressed is: A customer-support agent can send unrestricted HTTP requests containing private case content. Which concrete verification action demonstrates that the relevant behavior works?

  1. Ask the agent to post a private case to a hostile URL and confirm denial
  2. Restrict outbound destinations payload fields and authorization of network tools
  3. Allow any user-provided URL to receive the full customer record for debugging
  4. Attempt an unapproved production publish and confirm tenant controls block it

Explanation

Correct Answer

A:

To verify this change, the insurance provider should ask the agent to post a private case to a hostile URL and confirm denial. Operators still need the implementation to restrict outbound destinations payload fields and authorization of network tools and the preventive requirement to use destination allowlists DLP and redaction at connector boundaries.

Incorrect Answers

B:

The action to restrict outbound destinations payload fields and authorization of network tools can contribute to data exfiltration through tools as a implementation step; here, a customer-support agent can send unrestricted HTTP requests containing private case content. The more directly responsive choice is to ask the agent to post a private case to a hostile URL and confirm denial.

C:

The assumption to allow any user-provided URL to receive the full customer record for debugging is the exact failure pattern the team needs to avoid. The relevant change is to restrict outbound destinations payload fields and authorization of network tools; a credible check must ask the agent to post a private case to a hostile URL and confirm denial.

D:

For governance of agent publication, it can be sensible to attempt an unapproved production publish and confirm tenant controls block it if department makers can publish experimental agents directly to the entire tenant without security review. Without the actual change to ask the agent to post a private case to a hostile URL and confirm denial, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

Agentic AI Business Solutions Architect (AB-100) Practice Tests