Question 310
A Foundry operations group separates infrastructure administrators from application developers. The immediate design goal is to require an authorized access administrator when changing who may grant project or resource roles. Which approach addresses that goal?
- Inference data-plane access
- Resource management permission
- Project access restriction
- Privileged role assignment review
Explanation
Correct Answer
D:Role-assignment changes are sensitive authorization operations. Standard resource editing privileges should not be assumed sufficient to delegate broad access.
Incorrect Answers
A:Data-plane access authorizes supported inference operations; it does not imply the caller can redeploy the underlying model; the outstanding need is to require an authorized access administrator when changing who may grant project or resource roles.
B:Management-plane roles govern resource configuration actions; they should be distinguished from rights to submit individual inference requests; this approach leaves the team unable to require an authorized access administrator when changing who may grant project or resource roles.
C:Narrow scope reduces exposure to other workloads; a broad subscription assignment can unintentionally grant visibility or management across projects; the needed capability is to require an authorized access administrator when changing who may grant project or resource roles.
Question 311
A Foundry operations group separates infrastructure administrators from application developers. The team is required to account for role assignment propagation after changing an application's access grant. Which option is the best fit?
- Inference data-plane access
- Least-privilege access test
- Permission propagation check
- Resource management permission
Explanation
Correct Answer
C:Authorization changes can take time to become effective across services. Retrying without checking token freshness and scope can misdiagnose a transient state.
Incorrect Answers
A:Data-plane access authorizes supported inference operations; it does not imply the caller can redeploy the underlying model; the requirement, by contrast, is to account for role assignment propagation after changing an application's access grant.
B:Positive and negative tests confirm the intended permission boundary; successful inference alone cannot prove that administrative access is sufficiently restricted; the team would still need to account for role assignment propagation after changing an application's access grant.
D:Management-plane roles govern resource configuration actions; they should be distinguished from rights to submit individual inference requests; the implementation under review needs to account for role assignment propagation after changing an application's access grant.
Question 312
A Foundry application must invoke its approved model but must not create, delete, or reconfigure model deployments. A reviewer also wants a practical negative test proving that the app's identity lacks deployment-administration ability. Which TWO measures provide the intended runtime permission and a verification of its least-privilege boundary? Choose TWO.
- Permission propagation check
- Project access restriction
- Privileged role assignment review
- Least-privilege access test
- Inference data-plane access
Explanation
Correct Answers
D:Positive and negative tests confirm the intended permission boundary; successful inference alone cannot prove that administrative access is sufficiently restricted. Here, the runtime identity can score requests without being granted infrastructure change permissions.
E:Data-plane access authorizes supported inference operations; it does not imply the caller can redeploy the underlying model. Here, the negative test confirms a deployment-deletion attempt is forbidden.
Incorrect Answers
A:Authorization changes can take time to become effective across services; retrying without checking token freshness and scope can misdiagnose a transient state; the immediate requirement instead concerns restricted model invocation and a demonstrated ban on deployment administration.
B:Narrow scope reduces exposure to other workloads; a broad subscription assignment can unintentionally grant visibility or management across projects; the system would still need a remedy for restricted model invocation and a demonstrated ban on deployment administration.
C:Role-assignment changes are sensitive authorization operations; standard resource editing privileges should not be assumed sufficient to delegate broad access; this mechanism leaves unresolved restricted model invocation and a demonstrated ban on deployment administration.
Question 313
A generative AI service uses search and storage while minimizing long-lived secrets. The immediate design goal is to let an application authenticate to a supported Azure AI Search resource through scoped identity access. Which approach addresses that goal?
- Managed identity to search
- Managed identity to storage
- Connection permission validation
- Vault-backed legacy secret
Explanation
Correct Answer
A:Identity-based search access separates the caller's permissions from shared administrative keys. The required index or data access role must be assigned.
Incorrect Answers
B:Storage data-plane RBAC can restrict blob access by identity and scope; resource control-plane roles do not automatically imply access to all blob contents; the outstanding need is to let an application authenticate to a supported Azure AI Search resource through scoped identity access.
C:A model can generate fluent text while its search identity lacks data access; authentication and RBAC tests isolate this failure from prompt design; this approach leaves the team unable to let an application authenticate to a supported Azure AI Search resource through scoped identity access.
D:A secret vault centralizes restricted retrieval and rotation; it remains necessary to grant the workload identity access to the specific secret; the needed capability is to let an application authenticate to a supported Azure AI Search resource through scoped identity access.
Question 314
A generative AI service uses search and storage while minimizing long-lived secrets. The team is required to allow agent data pipelines to read only approved Blob storage containers through role grants. Which option is the best fit?
- Managed identity to storage
- Vault-backed legacy secret
- Key rotation workflow
- Credential-free build process
Explanation
Correct Answer
A:Storage data-plane RBAC can restrict blob access by identity and scope. Resource control-plane roles do not automatically imply access to all blob contents.
Incorrect Answers
B:A secret vault centralizes restricted retrieval and rotation; it remains necessary to grant the workload identity access to the specific secret; the requirement, by contrast, is to allow agent data pipelines to read only approved Blob storage containers through role grants.
C:Rotating keys constrains exposure from leaked secrets; merely hiding the key in a new settings page does not invalidate its prior value; the team would still need to allow agent data pipelines to read only approved Blob storage containers through role grants.
D:Keeping credentials out of immutable artifacts reduces leakage and simplifies rotation; runtime identity or secure secret retrieval supplies authorization when needed; the specified outcome instead requires the team to allow agent data pipelines to read only approved Blob storage containers through role grants.
Question 315
A generative AI service uses search and storage while minimizing long-lived secrets. An operator is tasked to store an unavoidable external API credential in a protected secret service rather than an application repository. Which mechanism provides that capability?
- Credential-free build process
- Connection permission validation
- Managed identity to storage
- Key rotation workflow
- Vault-backed legacy secret
Explanation
Correct Answer
E:A secret vault centralizes restricted retrieval and rotation. It remains necessary to grant the workload identity access to the specific secret.
Incorrect Answers
A:Keeping credentials out of immutable artifacts reduces leakage and simplifies rotation; runtime identity or secure secret retrieval supplies authorization when needed; the specified outcome instead requires the team to store an unavoidable external API credential in a protected secret service rather than an application repository.
B:A model can generate fluent text while its search identity lacks data access; authentication and RBAC tests isolate this failure from prompt design; the implementation under review needs to store an unavoidable external API credential in a protected secret service rather than an application repository.
C:Storage data-plane RBAC can restrict blob access by identity and scope; resource control-plane roles do not automatically imply access to all blob contents; the requested task remains to store an unavoidable external API credential in a protected secret service rather than an application repository.
D:Rotating keys constrains exposure from leaked secrets; merely hiding the key in a new settings page does not invalidate its prior value; this alternative does not achieve the requirement to store an unavoidable external API credential in a protected secret service rather than an application repository.
Question 316
A generative AI service uses search and storage while minimizing long-lived secrets. The immediate design goal is to replace an exposed model API credential and refresh consuming applications without downtime. Which approach addresses that goal?
- Vault-backed legacy secret
- Key rotation workflow
- Managed identity to storage
- Connection permission validation
Explanation
Correct Answer
B:Rotating keys constrains exposure from leaked secrets. Merely hiding the key in a new settings page does not invalidate its prior value.
Incorrect Answers
A:A secret vault centralizes restricted retrieval and rotation; it remains necessary to grant the workload identity access to the specific secret; this alternative does not achieve the requirement to replace an exposed model API credential and refresh consuming applications without downtime.
C:Storage data-plane RBAC can restrict blob access by identity and scope; resource control-plane roles do not automatically imply access to all blob contents; the outstanding need is to replace an exposed model API credential and refresh consuming applications without downtime.
D:A model can generate fluent text while its search identity lacks data access; authentication and RBAC tests isolate this failure from prompt design; this approach leaves the team unable to replace an exposed model API credential and refresh consuming applications without downtime.
Question 317
A generative AI service uses search and storage while minimizing long-lived secrets. The team is required to build application artifacts without injecting production inference credentials into the container image. Which option is the best fit?
- Credential-free build process
- Key rotation workflow
- Vault-backed legacy secret
- Connection permission validation
Explanation
Correct Answer
A:Keeping credentials out of immutable artifacts reduces leakage and simplifies rotation. Runtime identity or secure secret retrieval supplies authorization when needed.
Incorrect Answers
B:Rotating keys constrains exposure from leaked secrets; merely hiding the key in a new settings page does not invalidate its prior value; the needed capability is to build application artifacts without injecting production inference credentials into the container image.
C:A secret vault centralizes restricted retrieval and rotation; it remains necessary to grant the workload identity access to the specific secret; the requirement, by contrast, is to build application artifacts without injecting production inference credentials into the container image.
D:A model can generate fluent text while its search identity lacks data access; authentication and RBAC tests isolate this failure from prompt design; the team would still need to build application artifacts without injecting production inference credentials into the container image.
Question 318
A Foundry agent can call its model and Azure AI Search, but an enrichment step retrieves no blobs from a restricted container. No long-lived account key is allowed, the model and search connections are healthy, and the team must verify the agent's effective access before altering prompts. Which TWO actions target that storage-access failure? Choose TWO.
- Managed identity to search
- Credential-free build process
- Connection permission validation
- Vault-backed legacy secret
- Managed identity to storage
Explanation
Correct Answers
C:A model can generate fluent text while its search identity lacks data access; authentication and RBAC tests isolate this failure from prompt design. Here, the diagnosis tests the agent's effective permissions to the storage dependency.
E:Storage data-plane RBAC can restrict blob access by identity and scope; resource control-plane roles do not automatically imply access to all blob contents. Here, blob access is performed through scoped identity authorization rather than a stored key.
Incorrect Answers
A:Identity-based search access separates the caller's permissions from shared administrative keys; the required index or data access role must be assigned; this alternative does not resolve agent-to-Blob identity access and its effective connection permissions.
B:Keeping credentials out of immutable artifacts reduces leakage and simplifies rotation; runtime identity or secure secret retrieval supplies authorization when needed; this mechanism leaves unresolved agent-to-Blob identity access and its effective connection permissions.
D:A secret vault centralizes restricted retrieval and rotation; it remains necessary to grant the workload identity access to the specific secret; the observed fault still involves agent-to-Blob identity access and its effective connection permissions.
Question 319
The security team needs to stop public access to generative AI model resources. The immediate design goal is to provide an approved private IP path for supported model endpoint calls from a corporate virtual network. Which approach addresses that goal?
- Private DNS zone association
- Network security rule review
- Disable public network ingress
- Private endpoint for model service
Explanation
Correct Answer
D:A private endpoint moves service ingress onto a private network path. Clients still need DNS resolution and authorized identity credentials.
Incorrect Answers
A:Correct private DNS integration directs clients to approved private IPs; a provisioned private endpoint without matching DNS may remain unreachable; the requested task remains to provide an approved private IP path for supported model endpoint calls from a corporate virtual network.
B:NSG and firewall rules control the network traffic path; they do not authenticate the inference payload or select a model deployment; this alternative does not achieve the requirement to provide an approved private IP path for supported model endpoint calls from a corporate virtual network.
C:Public network restrictions close the unwanted ingress route; creating an internal DNS record alone would not block callers using the original public path; the actual design criterion is to provide an approved private IP path for supported model endpoint calls from a corporate virtual network.
Question 320
The security team needs to stop public access to generative AI model resources. The team is required to ensure application clients cannot bypass the approved private path by calling the public service endpoint. Which option is the best fit?
- Disable public network ingress
- Private connectivity verification
- Private endpoint for model service
- Network security rule review
Explanation
Correct Answer
A:Public network restrictions close the unwanted ingress route. Creating an internal DNS record alone would not block callers using the original public path.
Incorrect Answers
B:Testing from the deployed network verifies the effective DNS and routing path; a successful request from an engineer's laptop proves little about an isolated application subnet; this approach leaves the team unable to ensure application clients cannot bypass the approved private path by calling the public service endpoint.
C:A private endpoint moves service ingress onto a private network path; clients still need DNS resolution and authorized identity credentials; the needed capability is to ensure application clients cannot bypass the approved private path by calling the public service endpoint.
D:NSG and firewall rules control the network traffic path; they do not authenticate the inference payload or select a model deployment; the requirement, by contrast, is to ensure application clients cannot bypass the approved private path by calling the public service endpoint.