100% Latest & Updated Microsoft AB-100 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 380
A model provider announces a security issue affecting one deployed model version. What is the most defensible intervention for this case?
- Review model provenance, license and integrity before approving deployment.
- Inventory affected deployments and perform controlled replacement with regression checks.
- Preserve source classification and apply output handling controls to retrieved fragments.
- Revoke exposed credentials and use secure secret management with minimum scoped access.
Explanation
Correct Answer
B:Known model vulnerabilities require traceable remediation of the actual deployed versions.
Incorrect Answers
A:Third-party model supply-chain checks reduce security and legal uncertainty, whereas known model vulnerabilities require traceable remediation of the actual deployed versions.
C:Security labels must follow sensitive content through the generated response path, but known model vulnerabilities require traceable remediation of the actual deployed versions.
D:Credential exposure can enable unauthorized inference and data access. Known model vulnerabilities require traceable remediation of the actual deployed versions.
Question 381
A general service agent needs only a shipping lookup but can enumerate unrelated finance tools through the same gateway. Which specific remediation should the business adopt?
- Maintain an agent inventory with data sources, tools, privileges and classification.
- Sandbox computer-use access to approved applications and session scope.
- Separate execution identities and privilege domains for the two agents.
- Expose an allowlisted minimal tool catalog per agent role.
Explanation
Correct Answer
D:Tool discovery itself should not grant irrelevant sensitive capabilities.
Incorrect Answers
A:An inventory enables risk review of the organization’s actual AI attack surface. Tool discovery itself should not grant irrelevant sensitive capabilities.
B:GUI automation needs the same least-privilege separation as API tools, but tool discovery itself should not grant irrelevant sensitive capabilities.
C:Security boundaries prevent compromise of the research agent from granting payment authority. Tool discovery itself should not grant irrelevant sensitive capabilities.
Question 382
A read-only research agent shares a runtime identity with a payment-executing agent. Which solution approach best meets the stated requirement?
- Separate execution identities and privilege domains for the two agents.
- Derive authorization from validated identity claims, never from prompt content.
- Exchange scoped task assertions rather than long-lived secrets in agent messages.
- Sandbox computer-use access to approved applications and session scope.
Explanation
Correct Answer
A:Security boundaries prevent compromise of the research agent from granting payment authority.
Incorrect Answers
B:Natural-language role claims are untrusted and cannot expand tool privileges. Security boundaries prevent compromise of the research agent from granting payment authority.
C:Delegation contracts should not propagate reusable high-value credentials; security boundaries prevent compromise of the research agent from granting payment authority.
D:GUI automation needs the same least-privilege separation as API tools, whereas security boundaries prevent compromise of the research agent from granting payment authority.
Question 383
An untrusted browser automation tool can interact with a logged-in finance window not required by its task. Which option most directly closes the documented gap?
- Expose an allowlisted minimal tool catalog per agent role.
- Sandbox computer-use access to approved applications and session scope.
- Exchange scoped task assertions rather than long-lived secrets in agent messages.
- Derive authorization from validated identity claims, never from prompt content.
Explanation
Correct Answer
B:GUI automation needs the same least-privilege separation as API tools.
Incorrect Answers
A:Tool discovery itself should not grant irrelevant sensitive capabilities. GUI automation needs the same least-privilege separation as API tools.
C:Delegation contracts should not propagate reusable high-value credentials, whereas gUI automation needs the same least-privilege separation as API tools.
D:Natural-language role claims are untrusted and cannot expand tool privileges. GUI automation needs the same least-privilege separation as API tools.
Question 384
A specialist agent receives a delegation request containing raw credentials from a coordinator. Which approach best satisfies the actual business constraint?
- Separate execution identities and privilege domains for the two agents.
- Expose an allowlisted minimal tool catalog per agent role.
- Exchange scoped task assertions rather than long-lived secrets in agent messages.
- Revoke exposed credentials and use secure secret management with minimum scoped access.
Explanation
Correct Answer
C:Delegation contracts should not propagate reusable high-value credentials.
Incorrect Answers
A:Security boundaries prevent compromise of the research agent from granting payment authority; delegation contracts should not propagate reusable high-value credentials.
B:Tool discovery itself should not grant irrelevant sensitive capabilities, whereas delegation contracts should not propagate reusable high-value credentials.
D:Credential exposure can enable unauthorized inference and data access, but delegation contracts should not propagate reusable high-value credentials.
Question 385
Departments launch agents without assigned owners for business outcomes or security risk. Which design decision should the architect prioritize?
- Require named business owner, technical owner and risk accountability for each agent.
- Enforce data policies and connector governance before broad deployment.
- Separate execution identities and privilege domains for the two agents.
- Maintain an agent inventory with data sources, tools, privileges and classification.
Explanation
Correct Answer
A:Ownership aligns agents with responsible control and operating decisions.
Incorrect Answers
B:Approved connector boundaries prevent uncontrolled external data movement. Ownership aligns agents with responsible control and operating decisions.
C:Security boundaries prevent compromise of the research agent from granting payment authority, but ownership aligns agents with responsible control and operating decisions.
D:An inventory enables risk review of the organization’s actual AI attack surface. Ownership aligns agents with responsible control and operating decisions.
Question 386
A central team cannot see which agents are connected to regulated data or production tools. Which controlled change is most appropriate for this situation?
- Schedule policy and access revalidation against documented agent scope.
- Maintain an agent inventory with data sources, tools, privileges and classification.
- Assess and constrain data flows across all integrated external services.
- Enforce data policies and connector governance before broad deployment.
Explanation
Correct Answer
B:An inventory enables risk review of the organization’s actual AI attack surface.
Incorrect Answers
A:Ongoing review detects changes in risk after initial release approval. An inventory enables risk review of the organization’s actual AI attack surface.
C:Cross-service tool calls can cause regulated transfers even when core hosting is compliant. An inventory enables risk review of the organization’s actual AI attack surface.
D:Approved connector boundaries prevent uncontrolled external data movement; an inventory enables risk review of the organization’s actual AI attack surface.
Question 387
Makers frequently deploy custom connectors without review of outbound data destinations. Which design choice targets the observed problem?
- Require named business owner, technical owner and risk accountability for each agent.
- Allowlist reachable destinations and validate URLs before tool execution.
- Enforce data policies and connector governance before broad deployment.
- Schedule policy and access revalidation against documented agent scope.
Explanation
Correct Answer
C:Approved connector boundaries prevent uncontrolled external data movement.
Incorrect Answers
A:Ownership aligns agents with responsible control and operating decisions, but approved connector boundaries prevent uncontrolled external data movement.
B:Network and URL restrictions limit server-side request forgery and data exfiltration. Approved connector boundaries prevent uncontrolled external data movement.
D:Ongoing review detects changes in risk after initial release approval, whereas approved connector boundaries prevent uncontrolled external data movement.
Question 388
Leadership requests periodic evidence that active agents still match their approved operating purpose. Which concrete solution decision should the team make?
- Maintain an agent inventory with data sources, tools, privileges and classification.
- Require named business owner, technical owner and risk accountability for each agent.
- Apply location and minimization controls to telemetry and audit storage.
- Schedule policy and access revalidation against documented agent scope.
Explanation
Correct Answer
D:Ongoing review detects changes in risk after initial release approval.
Incorrect Answers
A:An inventory enables risk review of the organization’s actual AI attack surface. Ongoing review detects changes in risk after initial release approval.
B:Ownership aligns agents with responsible control and operating decisions; ongoing review detects changes in risk after initial release approval.
C:Observability data may contain protected information and require the same governance review, whereas ongoing review detects changes in risk after initial release approval.
Question 389
A service agent can be manipulated to issue a refund using an attacker-written customer message. Which recommendation addresses the identified failure most directly?
- Use a narrowly scoped service identity appropriate for the low-risk lookup.
- Threat-model the tool-action path and test prompt injection with server-side authorization.
- Evaluate retrieval poisoning risk and apply source trust/approval controls.
- Audit citation targets and source access controls for unintended disclosure.
Explanation
Correct Answer
B:A prompt defense alone is insufficient without transactional access enforcement.
Incorrect Answers
A:A bounded service credential can fit a shared public lookup when it grants no sensitive rights; a prompt defense alone is insufficient without transactional access enforcement.
C:External content can be altered to influence agent answers and tool choices. A prompt defense alone is insufficient without transactional access enforcement.
D:Even accurate citations can leak restricted locations or contents, but a prompt defense alone is insufficient without transactional access enforcement.
Question 390
A healthcare agent could expose private data via citations that link to accessible public caches. What should the solution architect recommend for this specific need?
- Evaluate retrieval poisoning risk and apply source trust/approval controls.
- Test inclusive interaction and provide accessible alternative channels.
- Audit citation targets and source access controls for unintended disclosure.
- Use authoritative retrieval, citation verification and escalation for unsupported legal claims.
Explanation
Correct Answer
C:Even accurate citations can leak restricted locations or contents.
Incorrect Answers
A:External content can be altered to influence agent answers and tool choices, whereas even accurate citations can leak restricted locations or contents.
B:Accessible design is a separate responsible-AI criterion from factual correctness. Even accurate citations can leak restricted locations or contents.
D:Evidence constraints and review address harmful fabricated citations. Even accurate citations can leak restricted locations or contents.
Agentic AI Business Solutions Architect (AB-100) Practice Tests