Microsoft AI-300 Practice Test 97
Question 960
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A managed online deployment repeatedly restarts because its startup script throws a missing-file exception before the liveness probe can succeed. The operator must safeguard training reproducibility and serving continuity. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results.
- Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings.
- Inspect request volume, concurrency, endpoint latency graphs, job timing and batch-friendly data path availability.
- Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment.
Explanation
Correct Answer
A:Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results. This is the most relevant evidence for endpoint deployment health probe fails before model initialization. Successful model registration does not guarantee scoring container initialization; startup dependencies must be validated in deployed environment. The matching controlled remediation is: Fix the image or artifact mount path and ensure init() loads the registered model before promoting the deployment. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A managed….
Incorrect Answers
B:Inspect evaluation artifact content, privacy classification, report permissions, explanation samples and downstream dashboard export settings. This would investigate model evaluation exposes sensitive examples in explainability report rather than the documented endpoint deployment health probe fails before model initialization issue. The relevant investigation is: Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results.
C:Inspect request volume, concurrency, endpoint latency graphs, job timing and batch-friendly data path availability. This would investigate overnight bulk scoring saturates interactive online endpoint rather than the documented endpoint deployment health probe fails before model initialization issue. The relevant investigation is: Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results.
D:Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. This would investigate canary rollout sends all traffic to unvalidated candidate rather than the documented endpoint deployment health probe fails before model initialization issue. The relevant investigation is: Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A managed….
Question 961
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A reusable feature-engineering component emits a URI folder while a downstream training step declares a tabular MLTable input, causing type validation to fail. The team must preserve approved access, model provenance and private network constraints. What should the engineer examine first to confirm the specific AI-300 workload issue?
- Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics.
- Compare environment image digest, Conda or pip dependencies, Docker build logs, locally resolved versions and job import traceback.
- Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies.
- Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error.
Explanation
Correct Answer
D:Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. This is the most relevant evidence for pipeline component output type mismatches consumer. Azure ML pipeline components have explicit typed ports; connecting incompatible artifact interfaces does not automatically transform the data. The matching controlled remediation is: Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema. This matters here because a reusable feature-engineering component emits a URI folder while a downstream training step declares a tabular….
Incorrect Answers
A:Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics. This would investigate aml managed network blocks model registry egress rather than the documented pipeline component output type mismatches consumer issue. The relevant investigation is: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error.
B:Compare environment image digest, Conda or pip dependencies, Docker build logs, locally resolved versions and job import traceback. This would investigate training environment lacks pinned package versions rather than the documented pipeline component output type mismatches consumer issue. The relevant investigation is: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error.
C:Check the component's registry name and version, target workspace identity, registry read permissions and referenced environment dependencies. This would investigate registry model asset not available to second workspace rather than the documented pipeline component output type mismatches consumer issue. The relevant investigation is: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. This matters here because a reusable feature-engineering component emits a URI folder while a downstream training step declares a tabular….
Question 962
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A training component references an image in a private Azure Container Registry but the job's identity lacks image pull privileges, so workers fail before Python starts. The team must preserve approved access, model provenance and private network constraints. What measured outcome validates the solution while preserving the stated constraints?
- Production resources appear only in the authorized scope and rerunning the script does not create duplicate sandbox resources.
- The job runs with a traceable environment digest and no surprise dependency upgrade on repeated execution.
- The pipeline resolves component dependencies and the training step reads the expected typed asset without implicit file guessing.
- Compute nodes pull the exact authorized image and run the component without granting registry push permissions.
Explanation
Correct Answer
D:Compute nodes pull the exact authorized image and run the component without granting registry push permissions. Container image pulls are authorized independently of Azure ML job submission; model training cannot start if its execution image cannot be obtained. a direct corrective action is to give the approved compute or workspace identity the necessary ACR pull access and use a validated immutable image reference. This matters here because a training component references an image in a private Azure Container Registry but the job's identity lacks image….
Incorrect Answers
A:Production resources appear only in the authorized scope and rerunning the script does not create duplicate sandbox resources. That addresses a different configuration issue rather than the main requirement. Verify instead: Compute nodes pull the exact authorized image and run the component without granting registry push permissions.
B:The job runs with a traceable environment digest and no surprise dependency upgrade on repeated execution. That addresses a different configuration issue rather than the main requirement. Verify instead: Compute nodes pull the exact authorized image and run the component without granting registry push permissions.
C:The pipeline resolves component dependencies and the training step reads the expected typed asset without implicit file guessing. That addresses a different configuration issue rather than the main requirement. Verify instead: Compute nodes pull the exact authorized image and run the component without granting registry push permissions. This matters here because a training component references an image in a private Azure Container Registry but the job's identity lacks image….
Question 963
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A loan-classification model meets aggregate accuracy targets but its acceptance review never checks false-negative rates by relevant customer cohorts, obscuring a material disparity. The operator must safeguard training reproducibility and serving continuity. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model.
- Inspect point-in-time feature joins, feature update timestamps, event time, training label time and registered retrieval specification.
- Restrict access, redact or aggregate sensitive examples and publish a privacy-reviewed evaluation artifact for authorized reviewers.
- Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
- Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration.
Explanation
Correct Answers
A:Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model. This targets responsible ai evaluation omits demographic cohort slices. Aggregate predictive quality does not establish responsible use across affected groups; evaluation must examine relevant harm and fairness dimensions. Verify the failure through: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. Validate success with: Promotion evidence reports acceptable cohort outcomes or blocks release when a disparity breaches approved fairness criteria. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A….
D:Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. This is the most relevant evidence for responsible ai evaluation omits demographic cohort slices. Aggregate predictive quality does not establish responsible use across affected groups; evaluation must examine relevant harm and fairness dimensions. The matching controlled remediation is: Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A….
Incorrect Answers
B:Inspect point-in-time feature joins, feature update timestamps, event time, training label time and registered retrieval specification. This is associated with online feature retrieval pulls values from after prediction time, not the essential evidence-and-fix pair for responsible ai evaluation omits demographic cohort slices. First examine: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. Then correct: Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model.
C:Restrict access, redact or aggregate sensitive examples and publish a privacy-reviewed evaluation artifact for authorized reviewers. This is associated with model evaluation exposes sensitive examples in explainability report, not the essential evidence-and-fix pair for responsible ai evaluation omits demographic cohort slices. First examine: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. Then correct: Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model.
E:Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration. This is associated with distributed job workers cannot communicate over required network path, not the essential evidence-and-fix pair for responsible ai evaluation omits demographic cohort slices. First examine: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. Then correct: Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model.
Question 964
During a regulated Azure ML workspace infrastructure review, the following condition occurs: An Azure ML pipeline consistently fails to read input blobs after a storage-account key rotation because its registered datastore uses the old key. The team must preserve approved access, model provenance and private network constraints. What approved correction resolves the failure without modifying unrelated healthy services?
- Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
- Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
- Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
- Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
Explanation
Correct Answer
C:Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management. This targets datastore authentication relies on expiring account key. Datastores provide connection configuration but do not bypass storage authentication; credential rotation must be propagated or identity access designed correctly. Verify the failure through: Inspect the datastore authentication mode, storage key rotation event, job input failure, and available identity-based storage permission assignments. Validate success with: A rerun reads the expected immutable input without hard-coded expired keys in notebooks or repository files.
Incorrect Answers
A:Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately. This could address azure ml workspace deployed in the wrong subscription, but it does not solve the specified datastore authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
B: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 authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
D:Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply. This could address iac workflow lacks repeatable infrastructure validation, but it does not solve the specified datastore authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
Question 965
As an AI operations team checks production agent evaluation evidence, the following condition occurs: A retrieval assistant generates fluent answers citing unrelated documents, yet the release dashboard declares success based solely on readability. The workflow must produce measurable quality, safety and performance evidence. Which acceptance test shows the AI operations defect is corrected?
- Unsupported claims fail the release gate even when the answers are grammatically smooth.
- Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately.
- Evaluation scores correspond to the intended example queries and reference answers without null or swapped fields.
- The service meets per-minute token and tail-latency targets under representative large-context requests.
Explanation
Correct Answer
A:Unsupported claims fail the release gate even when the answers are grammatically smooth. Readability and factual support are distinct dimensions of GenAI output quality; polished prose is not evidence that retrieved sources justify a claim. a direct corrective action is to add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency. This matters here because as an AI operations team checks production agent evaluation evidence, the following condition occurs: A retrieval….
Incorrect Answers
B:Historical baselines can be reproduced with the same evaluator version and new rubric results are labeled separately. That addresses a different configuration issue rather than the main requirement. Verify instead: Unsupported claims fail the release gate even when the answers are grammatically smooth. This matters here because as an AI operations team checks production agent evaluation evidence, the following condition occurs: A retrieval….
C: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: Unsupported claims fail the release gate even when the answers are grammatically smooth.
D:The service meets per-minute token and tail-latency targets under representative large-context requests. That addresses a different configuration issue rather than the main requirement. Verify instead: Unsupported claims fail the release gate even when the answers are grammatically smooth.
Question 966
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: An Azure ML pipeline consistently fails to read input blobs after a storage-account key rotation because its registered datastore uses the old key. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which production-safe remediation should the engineer apply to this AI workflow?
- Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
- Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
- Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances.
- Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
Explanation
Correct Answer
D:Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management. This targets datastore authentication relies on expiring account key. Datastores provide connection configuration but do not bypass storage authentication; credential rotation must be propagated or identity access designed correctly. Verify the failure through: Inspect the datastore authentication mode, storage key rotation event, job input failure, and available identity-based storage permission assignments. Validate success with: A rerun reads the expected immutable input without hard-coded expired keys in notebooks or repository files. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: An Azure ML….
Incorrect Answers
A:Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply. This could address iac workflow lacks repeatable infrastructure validation, but it does not solve the specified datastore authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: An Azure ML….
B: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 authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: An Azure ML….
C:Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances. This could address interactive compute instances remain running while unused, but it does not solve the specified datastore authentication relies on expiring account key failure. The matching remediation is: Replace the stale credential with a supported workspace identity-based datastore access path or rotate the authorized credential through secrets management.
Question 967
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A loan-classification model meets aggregate accuracy targets but its acceptance review never checks false-negative rates by relevant customer cohorts, obscuring a material disparity. The resolution must preserve quality gates, documented artifacts and rollback options. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Inspect point-in-time feature joins, feature update timestamps, event time, training label time and registered retrieval specification.
- Inspect registry model lifecycle status, alias assignments, endpoint references, deployment logs and retirement approvals.
- Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
- Examine sweep primary_metric, goal direction, logged metric names, candidate ranking and the held-out evaluation calculation.
Explanation
Correct Answer
C:Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. This is the most relevant evidence for responsible ai evaluation omits demographic cohort slices. Aggregate predictive quality does not establish responsible use across affected groups; evaluation must examine relevant harm and fairness dimensions. The matching controlled remediation is: Add appropriate privacy-respecting subgroup performance and fairness evaluation and document mitigation before promoting the model. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A….
Incorrect Answers
A:Inspect point-in-time feature joins, feature update timestamps, event time, training label time and registered retrieval specification. This would investigate online feature retrieval pulls values from after prediction time rather than the documented responsible ai evaluation omits demographic cohort slices issue. The relevant investigation is: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
B:Inspect registry model lifecycle status, alias assignments, endpoint references, deployment logs and retirement approvals. This would investigate retired model kept active as discoverable deployment candidate rather than the documented responsible ai evaluation omits demographic cohort slices issue. The relevant investigation is: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
D:Examine sweep primary_metric, goal direction, logged metric names, candidate ranking and the held-out evaluation calculation. This would investigate hyperparameter sweep minimizes the wrong objective rather than the documented responsible ai evaluation omits demographic cohort slices issue. The relevant investigation is: Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
Question 968
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A model platform pipeline applies Bicep changes immediately and breaks a production workspace because it never checks syntax, resource drift or the expected scope. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which TWO complementary choices identify the issue and correct it? Select TWO.
- Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals.
- Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics.
- Configure GitHub OIDC federated identity with a scoped Azure service principal and replace the long-lived secret in the deployment workflow.
- Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
- Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
Explanation
Correct Answers
A:Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. This is the most relevant evidence for iac workflow lacks repeatable infrastructure validation. Infrastructure automation should detect configuration drift and unexpected destructive changes before a production deployment proceeds. The matching controlled remediation is: Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A model platform….
E:Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply. This targets iac workflow lacks repeatable infrastructure validation. Infrastructure automation should detect configuration drift and unexpected destructive changes before a production deployment proceeds. Verify the failure through: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. Validate success with: Invalid or overbroad infrastructure changes fail validation and intended workspace updates deploy reproducibly after review. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A model platform….
Incorrect Answers
B:Examine workspace managed network isolation mode, outbound rules, registry destination, DNS and job network diagnostics. This is associated with aml managed network blocks model registry egress, not the essential evidence-and-fix pair for iac workflow lacks repeatable infrastructure validation. First examine: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. Then correct: Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
C:Configure GitHub OIDC federated identity with a scoped Azure service principal and replace the long-lived secret in the deployment workflow. This is associated with github action uses persistent azure client secret, not the essential evidence-and-fix pair for iac workflow lacks repeatable infrastructure validation. First examine: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. Then correct: Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
D:Update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access. This is associated with datastore uri selects the incorrect storage container, not the essential evidence-and-fix pair for iac workflow lacks repeatable infrastructure validation. First examine: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. Then correct: Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
Question 969
While preparing a secure Foundry model or agent release, the following condition occurs: A high-volume assistant deployment repeatedly hits throughput saturation during business-hour spikes because its provisioned throughput units were selected from average rather than peak token load. The remediation must meet enterprise access and reliability requirements without a broad bypass. What should the engineer examine first to confirm the specific AI-300 workload issue?
- Compare real tokenized input sizes, selected model version limits, deployment configuration and representative failure logs.
- Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
- Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
- Inspect deployed model modalities, model card capabilities, payload content types, supported API version and failed inference responses.
Explanation
Correct Answer
B:Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles. This is the most relevant evidence for provisioned throughput capacity insufficient at peak traffic. Provisioned model throughput constrains admitted tokens and request demand; average daily request counts cannot establish a safe peak capacity allocation. The matching controlled remediation is: Right-size the supported provisioned capacity or apply controlled request pacing and overflow strategy within budget and region limits. This matters here because while preparing a secure Foundry model or agent release, the following condition occurs: A high-volume assistant….
Incorrect Answers
A:Compare real tokenized input sizes, selected model version limits, deployment configuration and representative failure logs. This would investigate foundation model selected without required context window rather than the documented provisioned throughput capacity insufficient at peak traffic issue. The relevant investigation is: Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
C: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 provisioned throughput capacity insufficient at peak traffic issue. The relevant investigation is: Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
D:Inspect deployed model modalities, model card capabilities, payload content types, supported API version and failed inference responses. This would investigate vision input sent to text-only foundation model rather than the documented provisioned throughput capacity insufficient at peak traffic issue. The relevant investigation is: Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
Question 970
During a Foundry generative AI quality and observability incident, the following condition occurs: A tool-enabled finance agent passes harmful-text screening but executes instructions inside a hostile retrieved web page because its safety suite contains no prompt-injection or tool-misuse cases. The fix must preserve evaluation validity and avoid unnecessary user-data collection. Which targeted Azure AI operations change best addresses the documented problem?
- Make required quality/safety criteria explicit CI gates and stop deployment when evaluated scores fail the approved tolerances.
- Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.
- Use relevant retrieval context, justified response limits and supported caching after verifying quality on evaluation examples.
- Include appropriate complete conversation histories and tool events in the supported agent evaluation workflow while preserving privacy protections.
Explanation
Correct Answer
B:Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution. This targets content safety evaluation omits prompt injection tests. Harmful-content detection alone does not test whether an agent obeys malicious instructions inserted into lower-trust retrieved material. Verify the failure through: Inspect safety evaluator coverage, adversarial retrieval examples, trusted instruction hierarchy, tool-action traces and current release criteria. Validate success with: An adversarial document cannot cause unapproved transactions while authorized research questions still succeed. This matters here because a tool-enabled finance agent passes harmful-text screening but executes instructions inside a hostile retrieved web….
Incorrect Answers
A:Make required quality/safety criteria explicit CI gates and stop deployment when evaluated scores fail the approved tolerances. This could address evaluation workflow fails to block degraded agent deployment, but it does not solve the specified content safety evaluation omits prompt injection tests failure. The matching remediation is: Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.
C:Use relevant retrieval context, justified response limits and supported caching after verifying quality on evaluation examples. This could address prompt truncation strategy increases generated-token spend, but it does not solve the specified content safety evaluation omits prompt injection tests failure. The matching remediation is: Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.
D:Include appropriate complete conversation histories and tool events in the supported agent evaluation workflow while preserving privacy protections. This could address multi-turn agent evaluation uses isolated last message only, but it does not solve the specified content safety evaluation omits prompt injection tests failure. The matching remediation is: Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.