Domain 5: Threat Detection & Response for AI

Microsoft SC-500 Security Copilot · 49 questions

• A security operations analyst is using Microsoft Sentinel to detect anomalous usage of Azure OpenAI Service. The analyst wants to create an analytics rule that alerts when the number of Azure OpenAI API calls from a single IP address exceeds 10,000 within one hour. Which Sentinel analytics rule type is most appropriate?
• A Microsoft Sentinel playbook must automatically isolate an Azure VM when Microsoft Defender for Endpoint detects that the VM has been used to exfiltrate data from an Azure OpenAI deployment. The playbook should trigger automatically when an incident is created. Which Sentinel feature enables automatic playbook execution?
• A threat hunter is investigating potential data exfiltration from Azure OpenAI Service. Azure OpenAI diagnostic logs are ingested into Microsoft Sentinel. The hunter wants to find all instances where the token count in a single completion response exceeded 10,000 tokens in the last 24 hours. Which KQL query correctly retrieves this data?
• An organization's security team has identified a pattern where threat actors are using compromised developer credentials to enumerate Azure OpenAI model deployments and extract prompt configurations. Which Microsoft Sentinel data connector and analytic approach best detects this reconnaissance activity?
• A security team uses Microsoft Sentinel to monitor AI-related threats. They need to create a hunting query to find users who interacted with Azure OpenAI but whose IP addresses are listed in a threat intelligence watchlist. Which Sentinel feature links the watchlist to the query?
• The security operations team receives a Microsoft Defender XDR alert indicating that an Azure OpenAI resource has been accessed using a token that originated from an IP address in a high-risk country. The team needs to contain the threat immediately. Which response action should be taken first?
• A security analyst is investigating a Microsoft Sentinel incident involving suspicious Azure OpenAI usage. The analyst wants to correlate this incident with related Microsoft Defender XDR alerts to build a complete attack timeline. Which Sentinel feature provides this cross-product correlation?
• A Microsoft Sentinel analytics rule fires an alert for a suspected Azure OpenAI abuse incident. The security team wants to ensure that entity information (the source IP address and the Azure OpenAI resource name) is automatically mapped and enriched in the incident so analysts can pivot to entity pages immediately. Which Sentinel analytics rule configuration enables this?
• A Microsoft Sentinel analytics rule for Azure OpenAI threats generates 500 false positive alerts per day for legitimate high-volume API usage from a known batch processing service. The security team wants to suppress these alerts without disabling the rule. Which Sentinel feature should be used?
• A security engineer needs to configure Microsoft Sentinel to ingest Azure OpenAI Service diagnostic logs for threat detection. What are the two required steps to enable this data flow?
• A Microsoft Sentinel analytics rule needs to detect when Azure OpenAI API calls are made using tokens from accounts that are simultaneously flagged as risky in Microsoft Entra ID Protection. Which Sentinel feature enables correlation across these two data sources?
• A Sentinel playbook is triggered when a high-severity Azure OpenAI abuse incident is created. The playbook must automatically disable the Azure OpenAI deployment that is being abused to stop ongoing data exfiltration. Which Logic Apps action accomplishes this containment?
• A Microsoft Sentinel incident involves a threat actor performing model extraction by repeatedly querying an Azure OpenAI deployment with carefully crafted inputs to reconstruct the model's behavior. The security team wants to map this attack to the MITRE ATT&CK framework. Which MITRE ATT&CK for Enterprise tactic best describes model extraction?
• An organization's SOC team wants to proactively hunt for signs that an AI model API key has been leaked and is being used from external IP addresses. Azure OpenAI diagnostic logs are in Sentinel. Which KQL hunting query approach identifies calls from IPs not seen in the previous 30 days?
• A security team is implementing a SOAR playbook to automatically enrich Microsoft Sentinel AI security incidents with threat intelligence. When a new incident is created involving an external IP address, the playbook should look up the IP in Microsoft Defender Threat Intelligence (MDTI) and add the result as a comment to the Sentinel incident. Which Logic Apps connector enables the MDTI lookup?
• A Microsoft Sentinel analytics rule needs to be created to detect when an Azure OpenAI API key stored in Azure Key Vault is accessed by a service principal that is not in a pre-approved list. The analyst has a Sentinel watchlist named 'ApprovedAIServicePrincipals' with the approved ObjectIds. Which KQL approach correctly implements this detection?
• A security engineer is building a Microsoft Sentinel hunting notebook to detect anomalous patterns in Azure OpenAI usage using machine learning-based time-series analysis. Which KQL function enables detecting statistical anomalies in the number of Azure OpenAI API calls per hour over the past 30 days?
• A Microsoft Sentinel analytics rule generates an incident when Azure OpenAI Service returns a very high number of completion tokens in a short time, potentially indicating data exfiltration. The incident response team needs to automatically notify the AI security team via Microsoft Teams when such an incident is created. Which component in the playbook sends the Teams notification?
• A security analyst is using Microsoft Sentinel's User and Entity Behavior Analytics (UEBA) to investigate an Azure AI service account that has been flagged as having anomalous activity. The UEBA engine has raised an 'Anomalous Azure resource access' insight for this service principal. Where in Sentinel can the analyst see the full UEBA profile for this entity including timeline, peer comparison, and risk score?
• An organization's Microsoft Sentinel deployment needs to detect when an Azure OpenAI model is being queried with system prompt override attempts (jailbreak patterns) in near real-time (latency under 1 minute). Which Sentinel analytics rule type provides the lowest detection latency?
• A security team is investigating a Microsoft Defender XDR incident where a device used for AI model development shows signs of compromise. The device has Azure ML SDK and Azure CLI installed. The team is concerned that the attacker may have exfiltrated Azure credentials from the development environment. Which Defender XDR investigation step directly identifies if cached Azure credentials were accessed?
• A security team wants to set up a custom threat intelligence feed of Azure OpenAI endpoint abuse indicators (malicious IP addresses that have been observed submitting jailbreak prompts at scale) and use it in Sentinel analytics rules. Which Sentinel feature is the most direct mechanism for ingesting this custom threat intelligence?
• A security team is conducting an incident response exercise for a scenario where an organization's AI model used for loan approval may have been manipulated through data poisoning to approve fraudulent applications. Which artifact should the incident response team prioritize collecting to determine if poisoning occurred?
• An organization uses Azure AI Content Safety and wants to receive real-time alerts in Microsoft Sentinel when Azure AI Content Safety detects and blocks high-severity hate content in their AI application. Azure AI Content Safety metrics and logs are sent to a Log Analytics workspace connected to Sentinel. Which Sentinel component generates these alerts?
• A Microsoft Sentinel investigation reveals that an attacker used a compromised service principal to call the Azure OpenAI management API and enumerate all model deployments, then used that information to target specific gpt-4 deployments for abuse. The analyst wants to investigate the full attack chain using Sentinel's investigation graph. Which Sentinel feature provides this visual attack chain analysis?
• A Microsoft Sentinel workspace receives alerts from Microsoft Defender for AI about anomalous usage of an Azure OpenAI endpoint. A security engineer wants to automate the response by adding the source IP to a watchlist for 24 hours. Which Microsoft Sentinel feature should be used to automate this response?
• A security analyst is threat hunting for potential model inversion attacks against an Azure OpenAI deployment. The analyst wants to find API calls where the request body contains an unusually large number of tokens compared to the average. Which Microsoft Sentinel query approach should be used?
• A security team is building a Microsoft Sentinel analytics rule to detect when an AI application service principal authenticates from an unusual geographic location. The rule should correlate Entra ID sign-in logs with a watchlist of expected service principal locations. Which KQL approach is correct?
• An organization deploys a SOAR playbook in Microsoft Sentinel triggered when a Defender for AI alert fires for suspicious Azure OpenAI usage. The playbook must automatically revoke the API key used in the suspicious request. Which Logic App connector action should be used?
• A security analyst is investigating a potential data exfiltration via an AI chatbot. The analyst suspects a user is using the chatbot to extract information from an internal knowledge base by crafting specific questions. Which Microsoft Sentinel hunting query would best identify this behavior?
• A company's Microsoft Sentinel workspace ingests Azure OpenAI diagnostic logs. The security team wants a scheduled analytics rule that fires when the same prompt is submitted more than 20 times in 1 hour from different user sessions, which could indicate automated probing. Which Microsoft Sentinel feature should be used to detect near-duplicate prompts?
• An Azure Machine Learning pipeline uses a managed online endpoint to serve a fraud detection model. The security team discovers that model prediction logs are being retained for only 30 days, insufficient for forensic investigations requiring 1-year retention. What is the most efficient solution?
• A Microsoft Sentinel analytics rule needs to detect when an Azure OpenAI model is being used to generate large volumes of similar requests in a short period, potentially indicating automated abuse. A security engineer writes a KQL query to detect this. Which KQL operator is most appropriate for identifying time-windowed frequency anomalies?
• A security engineer is building a Microsoft Sentinel playbook to respond to a Defender for AI alert about a potential data exfiltration via Azure OpenAI. The playbook must enrich the alert with information about the source IP reputation before escalating. Which Logic App connector action provides IP reputation data?
• A security team discovers that an Azure OpenAI deployment is experiencing a credential stuffing attack where attackers cycle through many API keys attempting to find valid ones. Which combination of controls in Azure API Management (APIM) and Microsoft Entra ID provides the most comprehensive protection?
• A security engineer needs to create a Microsoft Sentinel workbook that shows the geographic distribution of Azure OpenAI API calls over the past 7 days, highlighting calls from countries outside the expected usage regions. Which Microsoft Sentinel feature provides geographic visualization capabilities?
• An organization wants to implement automated threat hunting for AI systems using Microsoft Sentinel. The security team wants to schedule hunting queries to run weekly and automatically create incidents when the query returns results indicating potential AI model abuse. Which Microsoft Sentinel feature converts a hunting query into scheduled incident creation?
• A security engineer is analyzing Microsoft Sentinel incidents related to AI workloads. The engineer notices that many incidents are false positives because the Azure OpenAI diagnostic logs show legitimate batch processing jobs that spike token usage. Which Microsoft Sentinel feature allows the engineer to tune detection rules to reduce false positives from known legitimate behavior?
• A security team is responding to an incident where an attacker has stolen an Azure OpenAI API key and is making unauthorized API calls. The team has already regenerated the key. However, they need to understand what was done with the stolen key. Which data source provides the most detailed forensic information about the API calls made using the compromised key?
• An organization deploys Microsoft Sentinel to detect threats against their AI platform. The security team wants to create a detection for when an Azure OpenAI fine-tuned model is modified or deleted without proper change management. Which Microsoft Sentinel data source captures these events?
• A security engineer needs to implement continuous monitoring for configuration drift in Azure AI security controls. Specifically, the engineer needs to detect when Azure OpenAI resources have their public network access re-enabled after being configured as private-only. Which Azure service provides real-time drift detection and alerting?
• A security architect is reviewing the disaster recovery plan for an AI security monitoring system built on Microsoft Sentinel. The Sentinel workspace is in the East US region. If East US becomes unavailable, the team needs to continue ingesting AI security alerts. Which Sentinel feature supports this requirement?
• A security engineer is investigating a Microsoft Sentinel incident that correlates multiple low-severity alerts into a high-confidence incident via the Fusion detection engine. The incident involves an anomalous Azure OpenAI usage spike following a compromised service principal sign-in. The engineer wants to understand how Fusion determined these alerts are related. Where can the engineer review the Fusion correlation logic?
• A Microsoft Sentinel playbook is triggered when a Defender for AI alert fires for 'Sensitive data exposed in AI response.' The playbook must automatically post a notification to a Microsoft Teams channel with the alert details. Which Logic App action step is required for this Teams notification?
• A security engineer builds a Microsoft Sentinel analytics rule that queries AzureDiagnostics for Azure OpenAI API calls and joins with SigninLogs to identify cases where a user accesses Azure OpenAI within 5 minutes of a suspicious sign-in. The KQL query has performance issues with high query times. Which KQL optimization should be applied first?
• A security engineer is configuring Microsoft Sentinel to detect when Azure OpenAI API usage occurs outside of approved business hours (8 AM to 6 PM, Monday through Friday, UTC). The engineer writes a KQL query using AzureDiagnostics logs. Which KQL function correctly extracts the hour and day of week from TimeGenerated to filter business hours?
• A security engineer needs to configure Microsoft Sentinel to automatically close a Defender for AI incident as a false positive when the source is a known internal penetration testing team's IP addresses. The pen test team's IPs are stored in a Sentinel watchlist. Which Microsoft Sentinel feature handles incident disposition automation?
• A security engineer needs to configure Microsoft Sentinel to detect when an Azure Machine Learning model training job accesses data outside its approved dataset scope. The engineer wants to detect when the training job service principal reads data from a storage account container not in the approved list. Which data source in Sentinel captures this activity?
• An organization is building a Microsoft Sentinel analytics rule to detect potential training data exfiltration from an Azure Machine Learning workspace. The analyst wants to alert when a data scientist downloads more than 1 GB of data from the training data storage account in a single day. Which KQL approach correctly implements this volume-based detection?