Microsoft AI-103 Dumps
Microsoft AI-103 Practice Test 45
Question 440
During a regional deployment, a cloud developer must accept a change for agent function authorization. The issue to be addressed is: An agent correctly formats a delete request but lacks user authority. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the travel support provider should test delete calls from unauthorized and authorized sample principals. Operators still need the implementation to authorize each tool action against the requester and resource before execution and the preventive requirement to require operation-level access checks independent of language model text.
Incorrect Answers
Choosing to attempt an out-of-scope Azure resource update and confirm denial would direct effort toward agent least-privilege service id, whose characteristic issue is that a tool-using agent can modify subscription settings unrelated to its job. The relevant corrective or preventive action here remains to test delete calls from unauthorized and authorized sample principals, with its own validation criteria.
The action to authorize each tool action against the requester and resource before execution can contribute to agent function authorization as a implementation step; here, an agent correctly formats a delete request but lacks user authority. In the travel support provider acceptance review, trying to authorize each tool action against the requester and resource before execution does not test the exact agent function authorization condition where an agent correctly formats a delete request but lacks user authority; the relevant test is to test delete calls from unauthorized and authorized sample principals.
In the travel support provider acceptance review, trying to grant all agents a broad delete permission because they use JSON schemas does not test the exact agent function authorization condition where an agent correctly formats a delete request but lacks user authority; the relevant test is to test delete calls from unauthorized and authorized sample principals. Acceptance should instead demonstrate that the service can test delete calls from unauthorized and authorized sample principals, using the described failure as the test criterion.
The action to require operation-level access checks independent of language model text can contribute to agent function authorization as a ongoing preventive rule step; here, an agent correctly formats a delete request but lacks user authority. In the travel support provider acceptance review, trying to require operation-level access checks independent of language model text does not test the exact agent function authorization condition where an agent correctly formats a delete request but lacks user authority; the relevant test is to test delete calls from unauthorized and authorized sample principals.
Question 441
At a regional hospital, the site reliability engineer is preparing a reliability exercise. The concrete requirement is: Separate customers see one another's earlier agent preferences. Which design action BEST addresses the stated need?
Explanation
Correct Answer
The proposed implementation is to scope stored conversation state and memory retrieval to the right tenant; this matches the operational requirement that separate customers see one another's earlier agent preferences. The team can simulate parallel tenants and check that no history leaks. The team should also audit memory access and enforce tenant-scoped conversation keys.
Incorrect Answers
The action to simulate parallel tenants and check that no history leaks can contribute to conversation state isolation as a targeted acceptance check step; here, separate customers see one another's earlier agent preferences. Choosing to simulate parallel tenants and check that no history leaks cannot directly replace the specified action to scope stored conversation state and memory retrieval to the right tenant when separate customers see one another's earlier agent preferences.
Choosing to reuse one shared conversation ID across all tenants to simplify state cannot directly replace the specified action to scope stored conversation state and memory retrieval to the right tenant when separate customers see one another's earlier agent preferences. The actual implementation must scope stored conversation state and memory retrieval to the right tenant, because the stated service condition is that separate customers see one another's earlier agent preferences.
Choosing to model ordered stages with an explicit wait for approval before the final tool call would direct effort toward sequential approval workflow, whose characteristic issue is that a procurement agent must validate a request before seeking human purchase approval. The relevant corrective or preventive action here remains to scope stored conversation state and memory retrieval to the right tenant, with its own validation criteria.
Question 442
After a source update at a insurance platform, the team observes this failure: Separate customers see one another's earlier agent preferences. 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 reuse one shared conversation ID across all tenants to simplify state; it conflicts with the actual conversation state isolation behavior needed when separate customers see one another's earlier agent preferences. The insurance platform should instead scope stored conversation state and memory retrieval to the right tenant, followed by a check to simulate parallel tenants and check that no history leaks.
Incorrect Answers
The action to allow unlimited recursive tool calls because the goal remains unsolved addresses loop termination safeguard rather than the decisive conversation state isolation condition. The insurance platform should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to reuse one shared conversation ID across all tenants to simplify state.
The alternative to audit memory access and enforce tenant-scoped conversation keys is a ongoing preventive rule step rather than the faulty assumption being diagnosed. This insurance platform failure-analysis alternative of audit memory access and enforce tenant-scoped conversation keys does not diagnose the particular conversation state isolation defect where separate customers see one another's earlier agent preferences; the keyed mistake is reuse one shared conversation ID across all tenants to simplify state.
This approach relates to search tool integration: an agent needs current internal documentation to answer product questions. If the team chose it instead, the targeted check to simulate parallel tenants and check that no history leaks could still fail because the underlying required action is to reuse one shared conversation ID across all tenants to simplify state.
The alternative to scope stored conversation state and memory retrieval to the right tenant is a implementation step rather than the faulty assumption being diagnosed. This insurance platform failure-analysis alternative of scope stored conversation state and memory retrieval to the right tenant does not diagnose the particular conversation state isolation defect where separate customers see one another's earlier agent preferences; the keyed mistake is reuse one shared conversation ID across all tenants to simplify state.
The alternative to simulate parallel tenants and check that no history leaks is a targeted acceptance check step rather than the faulty assumption being diagnosed. This insurance platform failure-analysis alternative of simulate parallel tenants and check that no history leaks does not diagnose the particular conversation state isolation defect where separate customers see one another's earlier agent preferences; the keyed mistake is reuse one shared conversation ID across all tenants to simplify state.
Question 443
The media platform is responding to a cost review. A agent designer discovers this issue: Separate customers see one another's earlier agent preferences. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The direct repair is to separate session identifiers and revoke leaked cross-tenant cache entries. Once the condition is corrected, the media platform should simulate parallel tenants and check that no history leaks. To reduce recurrence, it should also audit memory access and enforce tenant-scoped conversation keys; those steps provide different recovery and prevention functions.
Incorrect Answers
During media platform remediation, the proposed step to reuse one shared conversation ID across all tenants to simplify state is not the direct restoration action for conversation state isolation when separate customers see one another's earlier agent preferences; the keyed repair is separate session identifiers and revoke leaked cross-tenant cache entries. After remediation the team must simulate parallel tenants and check that no history leaks; the action in this option does not itself establish that target recovery outcome.
The action to simulate parallel tenants and check that no history leaks can contribute to conversation state isolation as a targeted acceptance check step; here, separate customers see one another's earlier agent preferences. During media platform remediation, the proposed step to simulate parallel tenants and check that no history leaks is not the direct restoration action for conversation state isolation when separate customers see one another's earlier agent preferences; the keyed repair is separate session identifiers and revoke leaked cross-tenant cache entries.
This approach relates to specialist agent selection: a finance question is routed to a generic vision agent with unrelated tools. If the team chose it instead, the targeted check to simulate parallel tenants and check that no history leaks could still fail because the underlying required action is to separate session identifiers and revoke leaked cross-tenant cache entries.
Question 444
A responsible-AI reviewer at a customer support center wants the following issue to be prevented in later deployments: Separate customers see one another's earlier agent preferences. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
Standardizing the boundary to audit memory access and enforce tenant-scoped conversation keys is preventive because the same conversation state isolation failure could otherwise return: separate customers see one another's earlier agent preferences. For this customer support center, the service should also scope stored conversation state and memory retrieval to the right tenant, and tests should simulate parallel tenants and check that no history leaks so that the ongoing requirement is measurable.
Incorrect Answers
The action to separate session identifiers and revoke leaked cross-tenant cache entries can contribute to conversation state isolation as a corrective remediation step; here, separate customers see one another's earlier agent preferences. For prevention at the customer support center, the measure to separate session identifiers and revoke leaked cross-tenant cache entries is weaker or aimed at another issue than the conversation state isolation guardrail to audit memory access and enforce tenant-scoped conversation keys needed when separate customers see one another's earlier agent preferences.
The action to maintain regression tests for action success and forbidden behavior addresses agent evaluation data rather than the decisive conversation state isolation condition. The customer support center should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to audit memory access and enforce tenant-scoped conversation keys.
For prevention at the customer support center, the measure to reuse one shared conversation ID across all tenants to simplify state is weaker or aimed at another issue than the conversation state isolation guardrail to audit memory access and enforce tenant-scoped conversation keys needed when separate customers see one another's earlier agent preferences. The lasting guardrail for this specific threat is to audit memory access and enforce tenant-scoped conversation keys, not simply to complete an adjacent service task.
Question 445
During a source update, a solution architect must accept a change for conversation state isolation. The issue to be addressed is: Separate customers see one another's earlier agent preferences. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
The targeted acceptance test is to simulate parallel tenants and check that no history leaks. The design should first scope stored conversation state and memory retrieval to the right tenant. A robust manufacturing group deployment also needs to audit memory access and enforce tenant-scoped conversation keys; passing this check provides relevant evidence but does not certify every possible input or security risk.
Incorrect Answers
For tool output schema validation, it can be sensible to inject missing timestamps and malformed temperatures into fixtures if an external weather tool occasionally returns malformed values. Without the actual change to simulate parallel tenants and check that no history leaks, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
In the manufacturing group acceptance review, trying to reuse one shared conversation ID across all tenants to simplify state does not test the exact conversation state isolation condition where separate customers see one another's earlier agent preferences; the relevant test is to simulate parallel tenants and check that no history leaks. Acceptance should instead demonstrate that the service can simulate parallel tenants and check that no history leaks, using the described failure as the test criterion.
The action to scope stored conversation state and memory retrieval to the right tenant can contribute to conversation state isolation as a implementation step; here, separate customers see one another's earlier agent preferences. In the manufacturing group acceptance review, trying to scope stored conversation state and memory retrieval to the right tenant does not test the exact conversation state isolation condition where separate customers see one another's earlier agent preferences; the relevant test is to simulate parallel tenants and check that no history leaks.
Question 446
At a travel support provider, the machine-learning developer is preparing a security tabletop exercise. The concrete requirement is: A web-search result instructs an agent to leak its configured credentials. Which design action BEST addresses the stated need?
Explanation
Correct Answer
Choosing to treat tool output as lower-trust evidence and ignore embedded action commands directly targets the described tool result untrusted content gap: a web-search result instructs an agent to leak its configured credentials. As a lasting boundary, they also need to enforce action allowlists and adversarial tool-output tests, because an implementation by itself does not prove safe operation.
Incorrect Answers
Choosing to require a user or authorized approver confirmation before destructive tool execution would direct effort toward human approval for destructive action, whose characteristic issue is that an agent could delete a business folder when asked to tidy files. The relevant corrective or preventive action here remains to treat tool output as lower-trust evidence and ignore embedded action commands, with its own validation criteria.
Choosing to execute any instruction quoted in a retrieved web page as higher-level policy cannot directly replace the specified action to treat tool output as lower-trust evidence and ignore embedded action commands when a web-search result instructs an agent to leak its configured credentials. The actual implementation must treat tool output as lower-trust evidence and ignore embedded action commands, because the stated service condition is that a web-search result instructs an agent to leak its configured credentials.
The action to return a malicious test result and verify no unauthorized action can contribute to tool result untrusted content as a targeted acceptance check step; here, a web-search result instructs an agent to leak its configured credentials. Choosing to return a malicious test result and verify no unauthorized action cannot directly replace the specified action to treat tool output as lower-trust evidence and ignore embedded action commands when a web-search result instructs an agent to leak its configured credentials.
Question 447
After a support escalation at a scientific publisher, the team observes this failure: A web-search result instructs an agent to leak its configured credentials. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?
Explanation
Correct Answer
The faulty assumption is execute any instruction quoted in a retrieved web page as higher-level policy. A functioning alternative would be to treat tool output as lower-trust evidence and ignore embedded action commands, and the relevant validation would be to return a malicious test result and verify no unauthorized action.
Incorrect Answers
The alternative to return a malicious test result and verify no unauthorized action is a targeted acceptance check step rather than the faulty assumption being diagnosed. This scientific publisher failure-analysis alternative of return a malicious test result and verify no unauthorized action does not diagnose the particular tool result untrusted content defect where a web-search result instructs an agent to leak its configured credentials; the keyed mistake is execute any instruction quoted in a retrieved web page as higher-level policy.
The alternative to treat tool output as lower-trust evidence and ignore embedded action commands is a implementation step rather than the faulty assumption being diagnosed. This scientific publisher failure-analysis alternative of treat tool output as lower-trust evidence and ignore embedded action commands does not diagnose the particular tool result untrusted content defect where a web-search result instructs an agent to leak its configured credentials; the keyed mistake is execute any instruction quoted in a retrieved web page as higher-level policy.
The option to issue a fresh purchase request on every transient error without deduplication would normally serve a different idempotent agent actions requirement, where a payment agent retries after a network disconnect and creates a duplicate purchase. The alternative may be useful in its own context, but it does not directly supply execute any instruction quoted in a retrieved web page as higher-level policy for the stated decision.
Question 448
The nonprofit organization is responding to a service audit. A quality assurance lead discovers this issue: A web-search result instructs an agent to leak its configured credentials. What corrective step should the team take to restore the intended functionality most directly?
Explanation
Correct Answer
The appropriate remediation is to remove the malicious content and rerun with injection-resistant tool boundaries, because the failure concerns tool result untrusted content: a web-search result instructs an agent to leak its configured credentials. A meaningful verification should return a malicious test result and verify no unauthorized action, and the durable operating rule should enforce action allowlists and adversarial tool-output tests after the immediate fix.
Incorrect Answers
During nonprofit organization remediation, the proposed step to execute any instruction quoted in a retrieved web page as higher-level policy is not the direct restoration action for tool result untrusted content when a web-search result instructs an agent to leak its configured credentials; the keyed repair is remove the malicious content and rerun with injection-resistant tool boundaries. After remediation the team must return a malicious test result and verify no unauthorized action; the action in this option does not itself establish that target recovery outcome.
The action to return a malicious test result and verify no unauthorized action can contribute to tool result untrusted content as a targeted acceptance check step; here, a web-search result instructs an agent to leak its configured credentials. During nonprofit organization remediation, the proposed step to return a malicious test result and verify no unauthorized action is not the direct restoration action for tool result untrusted content when a web-search result instructs an agent to leak its configured credentials; the keyed repair is remove the malicious content and rerun with injection-resistant tool boundaries.
This approach relates to typed tool function: an agent invents API parameters while calling an order-management tool. If the team chose it instead, the targeted check to return a malicious test result and verify no unauthorized action could still fail because the underlying required action is to remove the malicious content and rerun with injection-resistant tool boundaries.
Question 449
A AI platform engineer at a municipal services team wants the following issue to be prevented in later deployments: A web-search result instructs an agent to leak its configured credentials. Which enduring control or engineering guardrail BEST prevents recurrence?
Explanation
Correct Answer
The sustained prevention measure is to enforce action allowlists and adversarial tool-output tests. The implementation can still use the separate action to treat tool output as lower-trust evidence and ignore embedded action commands, but a guardrail constrains future changes. The municipal services team can measure whether the rule works by arranging to return a malicious test result and verify no unauthorized action.
Incorrect Answers
For prevention at the municipal services team, the measure to execute any instruction quoted in a retrieved web page as higher-level policy is weaker or aimed at another issue than the tool result untrusted content guardrail to enforce action allowlists and adversarial tool-output tests needed when a web-search result instructs an agent to leak its configured credentials. The lasting guardrail for this specific threat is to enforce action allowlists and adversarial tool-output tests, not simply to complete an adjacent service task.
The action to define typed response contracts with error paths addresses tool output schema validation rather than the decisive tool result untrusted content condition. The municipal services team should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to enforce action allowlists and adversarial tool-output tests.
The action to remove the malicious content and rerun with injection-resistant tool boundaries can contribute to tool result untrusted content as a corrective remediation step; here, a web-search result instructs an agent to leak its configured credentials. For prevention at the municipal services team, the measure to remove the malicious content and rerun with injection-resistant tool boundaries is weaker or aimed at another issue than the tool result untrusted content guardrail to enforce action allowlists and adversarial tool-output tests needed when a web-search result instructs an agent to leak its configured credentials.
Question 450
During a support escalation, a compliance specialist must accept a change for tool result untrusted content. The issue to be addressed is: A web-search result instructs an agent to leak its configured credentials. Which concrete verification action demonstrates that the relevant behavior works?
Explanation
Correct Answer
To verify this change, the payment services team should return a malicious test result and verify no unauthorized action. Operators still need the implementation to treat tool output as lower-trust evidence and ignore embedded action commands and the preventive requirement to enforce action allowlists and adversarial tool-output tests.
Incorrect Answers
In the payment services team acceptance review, trying to execute any instruction quoted in a retrieved web page as higher-level policy does not test the exact tool result untrusted content condition where a web-search result instructs an agent to leak its configured credentials; the relevant test is to return a malicious test result and verify no unauthorized action. Acceptance should instead demonstrate that the service can return a malicious test result and verify no unauthorized action, using the described failure as the test criterion.
For sequential approval workflow, it can be sensible to simulate rejection and verify purchase does not execute if a procurement agent must validate a request before seeking human purchase approval. Without the actual change to return a malicious test result and verify no unauthorized action, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.
The action to treat tool output as lower-trust evidence and ignore embedded action commands can contribute to tool result untrusted content as a implementation step; here, a web-search result instructs an agent to leak its configured credentials. In the payment services team acceptance review, trying to treat tool output as lower-trust evidence and ignore embedded action commands does not test the exact tool result untrusted content condition where a web-search result instructs an agent to leak its configured credentials; the relevant test is to return a malicious test result and verify no unauthorized action.
Developing AI Apps and Agents on Azure (AI-103) Practice Tests
- Microsoft AI-103 Practice Test 43 - 10 Questions (Last Reviewed 10-Oct-2026)
- Microsoft AI-103 Practice Test 44 - 10 Questions (Last Reviewed 09-Oct-2026)
- Microsoft AI-103 Practice Test 45 - 10 Questions (Last Reviewed 08-Oct-2026)
- Microsoft AI-103 Practice Test 46 - 10 Questions (Last Reviewed 07-Oct-2026)
- Microsoft AI-103 Practice Test 47 - 10 Questions (Last Reviewed 06-Oct-2026)