SC-500 Questions 21-30: Compute and AI Security

SC-500 validates end-to-end security controls for cloud and AI workloads. This set covers Secure compute, which accounts for 20–25% of the official skills outline. The domain tests server protection, application platform security, and controls for AI workloads. These SC-500 compute and AI questions focus on choosing protections that satisfy a specific operational constraint.

This is the third set in our Cloud and AI Security Engineer Associate SC-500 series (Questions 21-30). It follows SC-500 Questions 11-20 on storage and network security. Every question is an original practice scenario mapped to the official exam objectives, not a copy of a real exam item.

Distinguish boot prevention from boot integrity monitoring, and client authentication from backend authentication. Disabling prompt evidence does not disable AI threat analysis.

Question 21: Browser-Based Server Administration

Administrators need RDP access to Azure VMs through a browser in the Azure portal. The target VMs must have no public IP addresses, and administrators must not install a VPN client or maintain a jump server. Which service meets these requirements with the least infrastructure management?

A) Azure VPN Gateway with a point-to-site client connection.

B) Azure Bastion with a browser-based portal connection.

C) Azure Firewall with a public RDP forwarding rule.

D) An Azure VM acting as a public RDP jump server.

👁 Reveal Answer

Correct Answer: B

Explanation: B provides managed RDP and SSH connectivity through the portal over TLS, using the target VM’s private IP address. It avoids a customer-managed jump server and additional client software for browser access. A requires the VPN client connection excluded by the scenario. C forwards a management port rather than providing the required browser session. D creates a server that administrators must maintain, directly violating the infrastructure-management constraint.

Reference: What is Azure Bastion?

Question 22: An Access Window Ends

An Azure VM uses Defender for Servers Plan 2 just-in-time access through an NSG. No competing rule allows its management port. An authorized administrator opens an approved access window and establishes an SSH session. The window expires while that session is active. What should the administrator expect without assuming that the policy forcibly signs users out?

A) The existing session closes, and all management credentials are revoked.

B) The temporary allowance remains until the administrator closes the session.

C) The existing session closes, but new connections retain the temporary allowance.

D) The NSG returns to its previous state, but established connections continue.

👁 Reveal Answer

Correct Answer: D

Explanation: D matches the documented expiration behavior. Defender for Cloud restores the NSG’s previous state after the approved period, but already established connections are not interrupted. A incorrectly adds session termination and credential revocation. B incorrectly keeps the temporary network allowance until logout. C reverses both behaviors: existing connections continue, while the temporary rule no longer admits new connections. JIT limits network exposure; it is not a forced-session-termination mechanism.

Reference: Understand just-in-time virtual machine access

Question 23: Reject an Untrusted Boot Loader

A supported Generation 2 Azure VM uses Trusted Launch. Its security team must prevent an operating system boot loader that is not signed by a trusted publisher from starting. The team wants a boot-time enforcement control, not just a later health report. Which feature directly meets this requirement?

A) Secure Boot validation in the VM’s platform firmware.

B) vTPM storage of the VM’s measured boot information.

C) Guest attestation reporting of the VM’s boot integrity.

D) Defender for Cloud assessment of the VM’s configuration.

👁 Reveal Answer

Correct Answer: A

Explanation: A validates boot components against trusted publisher signatures. If Secure Boot cannot authenticate the boot image’s signature, the VM fails to boot. B provides protected measurements and keys for attestation, rather than this signature-enforcement step. C enables reporting and validation of boot integrity but does not replace Secure Boot’s enforcement. D identifies configuration concerns; a recommendation or assessment alone does not prevent an untrusted boot loader from executing.

Reference: Trusted Launch for Azure VMs

Question 24: Complete Boot Integrity Monitoring

A supported Trusted Launch VM already has Secure Boot and vTPM enabled. Defender for Cloud recommends an additional component so it can proactively attest and monitor the VM’s boot integrity. The component is not installed. Which configuration change completes the documented setup?

A) Disable vTPM, then install the Guest Attestation extension.

B) Disable Secure Boot, then install the Guest Attestation extension.

C) Keep both enabled, then install the Guest Attestation extension.

D) Keep both enabled, but omit the Guest Attestation extension.

👁 Reveal Answer

Correct Answer: C

Explanation: C supplies the missing extension used for remote attestation and boot integrity monitoring. The vTPM stores protected measurements of the boot chain, while Secure Boot enforces trusted boot components. A removes the vTPM that supplies protected boot measurements. B disables Secure Boot; Microsoft documents that attestation fails when it is disabled. D leaves the required monitoring extension absent and therefore does not resolve the recommendation. Keep Secure Boot and vTPM enabled as part of the documented monitoring configuration.

Reference: Trusted Launch for Azure VMs

Question 25: Bring a Physical Server Under Azure Management

A company must keep a supported Linux physical server in its own datacenter. It wants the server represented by an Azure resource ID and included in a resource group so it can apply Azure Machine Configuration and extend security management. Which onboarding approach meets these requirements without migrating the workload?

A) Install only the Azure Monitor Agent for operating system logs.

B) Install only Defender for Endpoint for endpoint threat detection.

C) Register the server’s application as an enterprise application.

D) Install and connect the Azure Arc Connected Machine agent.

👁 Reveal Answer

Correct Answer: D

Explanation: D onboards the non-Azure server to Azure Arc-enabled servers. The connected machine becomes an Azure resource with a resource ID and resource-group membership, supporting machine configuration and integrated management. A collects monitoring data but does not perform this Arc onboarding. B provides endpoint protection but does not create the required Arc management representation. C registers an application identity rather than the physical server as an Arc-enabled machine. The server remains in the datacenter.

Reference: Azure Arc-enabled servers overview

Question 26: Add Agentless VM Scanning

An Azure subscription currently uses Defender for Servers Plan 1. The team wants to retain its endpoint detection and response integration and add the agentless VM scanning capability included in Defender for Servers. Supported VMs and the required scanning permissions are available. Which plan change directly provides both capabilities?

A) Keep Servers Plan 1 and disable its endpoint integration.

B) Upgrade the subscription to Defender for Servers Plan 2.

C) Replace Servers Plan 1 with only Defender for Storage.

D) Replace Servers Plan 1 with only Defender for Containers.

👁 Reveal Answer

Correct Answer: B

Explanation: B includes Plan 1’s endpoint detection and response capabilities and adds agentless machine scanning. Enabling the appropriate plan is separate from satisfying the scanning prerequisites, which the scenario already provides. A removes an integration the team needs and does not add Plan 2 scanning. C protects storage workloads rather than providing the requested server package. D protects container workloads rather than supplying the required combination of server EDR and agentless VM scanning.

Reference: Select a Defender for Servers plan

Question 27: Observe One Kubernetes Namespace

An AKS cluster uses managed Microsoft Entra integration and Azure RBAC for Kubernetes authorization. An operations group already has the credentials and connectivity needed to reach the API. It needs read-only access to pods and deployments in the payments namespace, but must not read Secrets or change resources. Which built-in role assignment provides least privilege?

A) Assign AKS RBAC Writer at the payments namespace scope.

B) Assign AKS RBAC Admin at the payments namespace scope.

C) Assign AKS RBAC Reader at the payments namespace scope.

D) Assign AKS RBAC Reader at the entire cluster scope.

👁 Reveal Answer

Correct Answer: C

Explanation: C uses the Azure Kubernetes Service RBAC Reader role at the narrow namespace scope. It allows viewing most namespace objects while excluding Secrets, roles, and role bindings. A permits changes and access to Secrets, exceeding the requirement. B adds namespace administration capabilities. D uses the correct read-only role but extends access across the cluster instead of limiting it to payments. Authentication and API reachability are already established, so the missing decision is authorization scope.

Reference: Use Microsoft Entra ID authorization for the Kubernetes API in AKS

Question 28: Reject Anonymous API Requests

An App Service REST API has a valid Microsoft Entra identity provider configuration. Its built-in authentication setting currently allows unauthenticated requests. Security policy requires anonymous callers to receive HTTP 401 before the application handles a request, rather than a browser sign-in redirect. Which configuration meets the policy without changing application code?

A) Require authentication and return HTTP 401 for unauthenticated requests.

B) Require authentication and redirect unauthenticated requests with HTTP 302.

C) Allow unauthenticated requests and keep the identity provider configured.

D) Allow unauthenticated requests and disable the authentication token store.

👁 Reveal Answer

Correct Answer: A

Explanation: A instructs App Service’s platform middleware to reject anonymous traffic with the specified status before it reaches application code. B rejects anonymous access but sends a browser redirect, not the required HTTP 401. C leaves anonymous requests available despite having a sign-in provider. D changes token storage without requiring authentication. Built-in authentication runs separately from the app, so this configuration does not require implementing a new authentication flow in the API.

Reference: Authentication and authorization in Azure App Service

Question 29: Authenticate the Gateway to Its Backend

An API Management managed gateway must call an Entra-protected backend using its own system-assigned managed identity, not the caller’s token. The identity is enabled and authorized at the backend. The target resource identifier is known. Which inbound policy obtains a backend access token without storing a client secret?

A) Use authentication-basic with a named username and password.

B) Use authentication-managed-identity with the backend resource identifier.

C) Use validate-jwt against the incoming caller’s bearer token.

D) Use set-header to forward the incoming caller’s bearer token.

👁 Reveal Answer

Correct Answer: B

Explanation: B obtains an Entra access token for the specified backend resource using the gateway’s managed identity and sets it in the Authorization header with the Bearer scheme. Omitting client-id selects the system-assigned identity on a managed gateway. A relies on stored credentials. C validates an incoming token rather than obtaining the gateway’s own token. D forwards the caller’s identity, violating the requirement. Ensure that the resulting token goes only to the intended trusted backend.

Reference: API Management authentication-managed-identity policy

Question 30: Reduce Prompt Content in Alerts

A supported Azure OpenAI text workload has Defender for AI Services threat protection enabled. The security team wants threat analysis to continue, but it does not want user prompt content exposed as evidence in alerts. The engineer disables user prompt evidence while leaving the plan enabled. Which result should the team expect?

A) Threat analysis continues, but prompt content is masked in alerts.

B) Threat analysis stops, but previously collected alert evidence remains visible.

C) Threat analysis continues, but all prompt content remains visible in alerts.

D) Threat analysis stops, but prompt evidence is replaced by image scanning.

👁 Reveal Answer

Correct Answer: A

Explanation: A is the documented behavior when user prompt evidence is disabled: Defender for Cloud continues analyzing prompts and responses for threat detection, but masks prompt content in alerts. B incorrectly equates evidence display with stopping threat analysis. C ignores the masking behavior of the disabled setting. D invents a change of scanning modality instead of recognizing an evidence-display control. This setting reduces content exposed in alerts; it does not mean that prompts are no longer processed for detection.

Reference: Enable threat protection for AI services

Study Tips for Compute and AI Security

• Separate prevention from evidence. Secure Boot enforces trusted signatures; vTPM and guest attestation support boot integrity validation.
• Read the access window precisely. JIT restores network rules after expiration but does not interrupt established connections.
• Keep identities distinct. Authenticating a caller does not obtain a gateway identity token for the backend.
• Limit evidence without assuming no processing. Disabling user prompt evidence masks alert content while AI threat analysis continues.

Keep Practicing for Cloud and AI Security Engineer

Use these SC-500 compute and AI questions to review the boundary between enforcement and monitoring. Revisit SC-500 Questions 11-20 for storage and network security. The next set will cover security posture and monitoring.