Microsoft AI-300 Practice Test 33
Question 320
While preparing a secure Foundry model or agent release, the following condition occurs: A developer changes a production agent prompt directly in the Foundry portal without committing it, so a later regression cannot be tied to the release history. The remediation must meet enterprise access and reliability requirements without a broad bypass. What measured outcome validates the solution while preserving the stated constraints?
- Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content.
- The workflow authenticates through a governed identity after old secrets are disabled and completes the intended evaluation without privileged keys in source control.
- The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset.
- Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft.
Explanation
Correct Answer
D:Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft. Prompt text is production code-like configuration; untracked edits undermine provenance, collaboration and controlled rollback. a direct corrective action is to store reviewed prompt templates and configuration in version control and deploy immutable approved revisions through the release workflow.
Incorrect Answers
A:Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content. That addresses a different configuration issue rather than the main requirement. Verify instead: Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft.
B:The workflow authenticates through a governed identity after old secrets are disabled and completes the intended evaluation without privileged keys in source control. That addresses a different configuration issue rather than the main requirement. Verify instead: Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft.
C:The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset. That addresses a different configuration issue rather than the main requirement. Verify instead: Every production prompt version resolves to a reviewed commit and a failed change can be rolled back without copying an unknown portal draft.
Question 321
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A GitHub deployment workload that only uploads models receives broad subscription Contributor rights, enabling it to modify unrelated application infrastructure. The team must preserve approved access, model provenance and private network constraints. Which acceptance test shows the AI operations defect is corrected?
- Authorized workspaces run the shared component while unauthorized subscriptions remain blocked from the registry.
- Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions.
- Idle costs fall outside approved usage windows and users can resume the instances for sanctioned experiments.
- The job reads the required training blobs and cannot write or list unrelated protected containers.
Explanation
Correct Answer
B:Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions. An Azure resource role assignment grants management-plane capabilities according to scope and cannot be justified solely by pipeline convenience. a direct corrective action is to scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
Incorrect Answers
A:Authorized workspaces run the shared component while unauthorized subscriptions remain blocked from the registry. That addresses a different configuration issue rather than the main requirement. Verify instead: Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions.
C:Idle costs fall outside approved usage windows and users can resume the instances for sanctioned experiments. That addresses a different configuration issue rather than the main requirement. Verify instead: Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions.
D:The job reads the required training blobs and cannot write or list unrelated protected containers. That addresses a different configuration issue rather than the main requirement. Verify instead: Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions.
Question 322
During a Foundry generative AI quality and observability incident, the following condition occurs: A team upgrades a custom evaluator prompt and compares this month's agent scores to last month's, even though the grading rubric and baseline model changed. The fix must preserve evaluation validity and avoid unnecessary user-data collection. Which observable postchange result proves the intended remediation succeeded?
- The evaluation reveals and measures failures that depend on follow-up context and the selected agent version improves those cases.
- Evaluation scores correspond to the intended example queries and reference answers without null or swapped fields.
- An induced tool timeout is attributable to the exact trace span and remediation can target the failing service.
- Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately.
Explanation
Correct Answer
D:Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately. Evaluation metrics are outputs of versioned assessment logic; changing a judge or rubric can change scores without any underlying agent change. a direct corrective action is to version the custom evaluator and recalibrate on a fixed reference set before using its scores for cross-version comparison. This matters here because a team upgrades a custom evaluator prompt and compares this month's agent scores to last month's, even though the….
Incorrect Answers
A:The evaluation reveals and measures failures that depend on follow-up context and the selected agent version improves those cases. That addresses a different configuration issue rather than the main requirement. Verify instead: Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately.
B:Evaluation scores correspond to the intended example queries and reference answers without null or swapped fields. That addresses a different configuration issue rather than the main requirement. Verify instead: Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately.
C:An induced tool timeout is attributable to the exact trace span and remediation can target the failing service. That addresses a different configuration issue rather than the main requirement. Verify instead: Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately.
Question 323
During a retrieval or model-customization performance review, the following condition occurs: An enterprise fine-tuned assistant passes its original launch test but performs poorly after new product codes appear, and no evaluation monitor samples the updated terminology. The change must keep citations, privacy and previously approved behavior intact. What measured outcome validates the solution while preserving the stated constraints?
- A disaster-recovery deployment loads the reviewed customized model with matching tokenizer and runtime without repeating unrecoverable training.
- Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence.
- The selected retrieval strategy meets groundedness and task-success gates rather than optimizing click-through in isolation.
- The approved version meets safety requirements on held-out adversarial tests and its model artifact is traceable to the training record.
Explanation
Correct Answer
B:Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence. Fine-tuned models can become stale as input distribution shifts; continuous quality monitoring is needed beyond deployment-day accuracy. a direct corrective action is to refresh a governed evaluation set with recent terminology and trigger controlled retraining only after measuring sustained quality regression. This matters here because an enterprise fine-tuned assistant passes its original launch test but performs poorly after new product codes….
Incorrect Answers
A:A disaster-recovery deployment loads the reviewed customized model with matching tokenizer and runtime without repeating unrecoverable training. That addresses a different configuration issue rather than the main requirement. Verify instead: Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence.
C:The selected retrieval strategy meets groundedness and task-success gates rather than optimizing click-through in isolation. That addresses a different configuration issue rather than the main requirement. Verify instead: Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence.
D:The approved version meets safety requirements on held-out adversarial tests and its model artifact is traceable to the training record. That addresses a different configuration issue rather than the main requirement. Verify instead: Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence. This matters here because an enterprise fine-tuned assistant passes its original launch test but performs poorly after new product codes….
Question 324
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A central model team registers a reusable component in a shared Azure ML registry, but application workspace users cannot resolve the component because registry access was never granted. The team must preserve approved access, model provenance and private network constraints. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Inspect MLTable path definitions, transformation instructions, sampled file headers, declared feature schema and job parsing errors.
- Compare compute-instance schedules, last interactive activity, active notebook kernels, managed identity dependencies and cost reports.
- Inspect component YAML for workspace IDs, hard-coded URIs, parameter definitions and actual target workspace data asset registrations.
- Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies.
Explanation
Correct Answer
D:Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies. This is the most relevant evidence for registry model asset not available to second workspace. Cross-workspace registries support reusing vetted assets but do not automatically authorize every workspace or resolve unknown versions. The matching controlled remediation is: Grant controlled registry consumption rights and use the fully qualified versioned registry asset reference in the target workspace.
Incorrect Answers
A:Inspect MLTable path definitions, transformation instructions, sampled file headers, declared feature schema and job parsing errors. This would investigate mltable data asset schema mismatches training code rather than the documented registry model asset not available to second workspace issue. The relevant investigation is: Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies.
B:Compare compute-instance schedules, last interactive activity, active notebook kernels, managed identity dependencies and cost reports. This would investigate interactive compute instances remain running while unused rather than the documented registry model asset not available to second workspace issue. The relevant investigation is: Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies.
C:Inspect component YAML for workspace IDs, hard-coded URIs, parameter definitions and actual target workspace data asset registrations. This would investigate reusable component hardcodes workspace-specific asset path rather than the documented registry model asset not available to second workspace issue. The relevant investigation is: Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies.
Question 325
While preparing a secure Foundry model or agent release, the following condition occurs: An API service rolls back prompt text after a deployment incident but leaves the incompatible new foundation-model version in place, so tool-call formatting errors persist. The remediation must meet enterprise access and reliability requirements without a broad bypass. What measured outcome validates the solution while preserving the stated constraints?
- The last stable response and tool-call contracts are recovered across representative transactions without rebuilding application data.
- Replaying approved test prompts yields compatible output contracts until the next reviewed version is promoted.
- The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected.
- The reviewed deployment succeeds without reusable secrets in committed files and old credentials no longer authorize calls.
Explanation
Correct Answer
A:The last stable response and tool-call contracts are recovered across representative transactions without rebuilding application data. A generative application behavior depends on versioned combinations of model, prompts and tools, not any single text template. a direct corrective action is to restore the complete compatible model, prompt and tool schema release manifest and retest its evaluated workflow.
Incorrect Answers
B:Replaying approved test prompts yields compatible output contracts until the next reviewed version is promoted. That addresses a different configuration issue rather than the main requirement. Verify instead: The last stable response and tool-call contracts are recovered across representative transactions without rebuilding application data.
C:The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected. That addresses a different configuration issue rather than the main requirement. Verify instead: The last stable response and tool-call contracts are recovered across representative transactions without rebuilding application data.
D:The reviewed deployment succeeds without reusable secrets in committed files and old credentials no longer authorize calls. That addresses a different configuration issue rather than the main requirement. Verify instead: The last stable response and tool-call contracts are recovered across representative transactions without rebuilding application data.
Question 326
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A training job reads a similarly named blob container containing last quarter's features because its Azure ML datastore URI targets the staging location rather than the approved production store. The team must preserve approved access, model provenance and private network constraints. Which production-safe remediation should the engineer apply to this AI workflow?
- Gate production deployment behind approved GitHub environment protection, protected refs and status checks while keeping test deployments separate.
- Link the authorized Key Vault or repair supported access to the intended secret store through the correct workspace deployment configuration.
- Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
- Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
Explanation
Correct Answer
D:Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access. This targets datastore uri selects the incorrect storage container. A datastore connection and data asset version identify where training inputs come from; paths must be validated independently from authentication success. Verify the failure through: Compare the datastore URI and backing storage container, recorded data asset path, job input mount path and expected data version checksum. Validate success with: The job consumes the intended dataset version and staging files are not silently mixed into production runs.
Incorrect Answers
A:Gate production deployment behind approved GitHub environment protection, protected refs and status checks while keeping test deployments separate. This could address github deployment pipeline omits secure environment approvals, but it does not solve the specified datastore uri selects the incorrect storage container failure. The matching remediation is: Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
B:Link the authorized Key Vault or repair supported access to the intended secret store through the correct workspace deployment configuration. This could address workspace dependencies point to unrelated secrets vault, but it does not solve the specified datastore uri selects the incorrect storage container failure. The matching remediation is: Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
C:Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant. This could address workspace contributor role granted to untrusted automation, but it does not solve the specified datastore uri selects the incorrect storage container failure. The matching remediation is: Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
Question 327
While preparing a secure Foundry model or agent release, the following condition occurs: A custom transformer model is deployed on managed compute but inference containers crash on startup because the image's tokenizer library is incompatible with the model artifact. The remediation must meet enterprise access and reliability requirements without a broad bypass. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
- Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest.
- Select an approved multimodal model or add a supported image-to-text preprocessing service before the text generation step.
- Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
- Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
Explanation
Correct Answers
A:Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment. This targets managed foundation model deployment lacks compatible runtime image. Managed inference still requires the model weights and runtime dependencies to be compatible; compute provisioning cannot fix a broken model-serving image. Verify the failure through: Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. Validate success with: Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs.
B:Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. This is the most relevant evidence for managed foundation model deployment lacks compatible runtime image. Managed inference still requires the model weights and runtime dependencies to be compatible; compute provisioning cannot fix a broken model-serving image. The matching controlled remediation is: Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
Incorrect Answers
C:Select an approved multimodal model or add a supported image-to-text preprocessing service before the text generation step. This is associated with vision input sent to text-only foundation model, not the essential evidence-and-fix pair for managed foundation model deployment lacks compatible runtime image. First examine: Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. Then correct: Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
D:Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles. This is associated with provisioned throughput capacity insufficient at peak traffic, not the essential evidence-and-fix pair for managed foundation model deployment lacks compatible runtime image. First examine: Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. Then correct: Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
E:Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration. This is associated with agent runtime managed identity cannot read retrieval data, not the essential evidence-and-fix pair for managed foundation model deployment lacks compatible runtime image. First examine: Examine model package runtime requirements, container startup traceback, tokenizer/config files and managed compute environment image digest. Then correct: Build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
Question 328
While optimizing a production RAG or fine-tuned model service, the following condition occurs: An enterprise assistant's new retrieval strategy raises link clicks but adds hallucinated recommendations, while the experiment reports only engagement as the success metric. The resulting model must meet held-out quality and security criteria. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Inspect before-and-after safety evaluations, fine-tune dataset sources, refusal behavior, deployment policy and model version lineage.
- Review A/B cohort assignment, groundedness samples, relevance labels, business success measures and unsupported-claim incidents.
- Compare query and index embedding model versions, vector dimensions, index schema, retrieval score distributions and migration logs.
- Check document effective dates, index filters, scoring profiles, vector/keyword rankings and available canonical document metadata.
Explanation
Correct Answer
B:Review A/B cohort assignment, groundedness samples, relevance labels, business success measures and unsupported-claim incidents. This is the most relevant evidence for rag a/b experiment judges only click-through not factuality. Interaction metrics do not establish answer accuracy; retrieval optimization needs labels and safety/quality measurements. The matching controlled remediation is: Evaluate both retrievers with relevance, factual support and business outcomes under stable randomized cohorts and explicit guardrails.
Incorrect Answers
A:Inspect before-and-after safety evaluations, fine-tune dataset sources, refusal behavior, deployment policy and model version lineage. This would investigate fine-tuned model loses safety boundary after customization rather than the documented rag a/b experiment judges only click-through not factuality issue. The relevant investigation is: Review A/B cohort assignment, groundedness samples, relevance labels, business success measures and unsupported-claim incidents.
C:Compare query and index embedding model versions, vector dimensions, index schema, retrieval score distributions and migration logs. This would investigate embedding model replaced without reindexing stored vectors rather than the documented rag a/b experiment judges only click-through not factuality issue. The relevant investigation is: Review A/B cohort assignment, groundedness samples, relevance labels, business success measures and unsupported-claim incidents.
D:Check document effective dates, index filters, scoring profiles, vector/keyword rankings and available canonical document metadata. This would investigate hybrid retrieval double-counts outdated documents rather than the documented rag a/b experiment judges only click-through not factuality issue. The relevant investigation is: Review A/B cohort assignment, groundedness samples, relevance labels, business success measures and unsupported-claim incidents.
Question 329
While preparing a secure Foundry model or agent release, the following condition occurs: A global customer-support agent project is provisioned in a region that does not offer the selected foundation model deployment type, preventing its evaluation pipeline from starting. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which production-safe remediation should the engineer apply to this AI workflow?
- Migrate the workflow to supported managed or federated identity, or rotate the credential through approved secret management and verify minimal access.
- Deploy the governed Foundry resource and project in a supported compliant region with an available model option.
- Separate trusted instructions from quoted retrieved context and enforce restricted tool permissions and injection-resilient prompt handling.
- Replace broad management roles with the minimum supported inference access permission and remove redundant elevated grants.
Explanation
Correct Answer
B:Deploy the governed Foundry resource and project in a supported compliant region with an available model option. This targets foundry project created in mismatched resource location. A Foundry project depends on its parent resource location and model/service availability; creating a project does not guarantee every model is deployable there. Verify the failure through: Inspect Foundry project and resource regions, model deployment regional availability, enterprise residency constraints and API error codes. Validate success with: The project invokes the approved model endpoint and meets required data-residency policy without an unapproved region fallback.
Incorrect Answers
A:Migrate the workflow to supported managed or federated identity, or rotate the credential through approved secret management and verify minimal access. This could address foundry service principal credential rotated without updating pipeline, but it does not solve the specified foundry project created in mismatched resource location failure. The matching remediation is: Deploy the governed Foundry resource and project in a supported compliant region with an available model option.
C:Separate trusted instructions from quoted retrieved context and enforce restricted tool permissions and injection-resilient prompt handling. This could address prompt includes untrusted retrieved instructions as system policy, but it does not solve the specified foundry project created in mismatched resource location failure. The matching remediation is: Deploy the governed Foundry resource and project in a supported compliant region with an available model option.
D:Replace broad management roles with the minimum supported inference access permission and remove redundant elevated grants. This could address foundry administrator role granted to consumer application, but it does not solve the specified foundry project created in mismatched resource location failure. The matching remediation is: Deploy the governed Foundry resource and project in a supported compliant region with an available model option.
Question 330
During a Microsoft Foundry generative AI deployment review, the following condition occurs: A GenAI application releases its Foundry project and calls a model name before that model deployment is created, causing intermittent first-run errors. The deployment must retain governed identities, endpoint contracts and auditable release history. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Compare project connection configuration, model deployment resource ID, endpoint URL, deployment name and invocation authorization.
- Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
- Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds.
- Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations.
Explanation
Correct Answer
D:Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. This is the most relevant evidence for foundry bicep rollout omits model deployment dependencies. Infrastructure-as-code can provision the platform consistently only when dependencies and deployment readiness are managed explicitly. The matching controlled remediation is: Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
Incorrect Answers
A:Compare project connection configuration, model deployment resource ID, endpoint URL, deployment name and invocation authorization. This would investigate foundry project links to wrong deployed model resource rather than the documented foundry bicep rollout omits model deployment dependencies issue. The relevant investigation is: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations.
B:Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. This would investigate foundry private endpoint dns resolves to public ip rather than the documented foundry bicep rollout omits model deployment dependencies issue. The relevant investigation is: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations.
C:Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds. This would investigate prompt variant comparison measures fluency not task success rather than the documented foundry bicep rollout omits model deployment dependencies issue. The relevant investigation is: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations.