AI Bot
AI-500: Security, Governance & Deployment Flashcards
Microsoft Multi-Agent AI Solutions Expert AI-500 Flashcards
Study our AI-500: Security, Governance & Deployment flashcards for the Microsoft Multi-Agent AI Solutions Expert AI-500 exam with 40+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Agent orchestration strategies | Use Kubernetes or Azure Container Apps for lifecycle management scaling health checks and resource isolation |
| Audit and compliance logging | Capture who accessed what when and why store logs in immutable storage and retain per policy |
| Blue green deployment advantage | Maintains two production environments to enable instant rollback and zero downtime switches |
| Canary deployment purpose | Gradually route a small percentage of traffic to a new model version to validate behavior before full rollout |
| CI CD pipeline steps for agents | Source build test model validation security scans containerize and deploy with automated tests and rollbacks |
| Cost optimization tactics for inference | Use autoscaling right sized VMs containers GPU sharing and batch processing to reduce cost |
| Data minimization best practice | Collect only required data apply retention limits and anonymize or pseudonymize sensitive fields |
| Data residency and compliance | Store and process data in required geographic regions and document controls to meet local regulations |
| Differential privacy basic concept | A mathematical technique to add controlled noise to queries to protect individual data while preserving aggregate utility |
| Encryption at rest vs in transit on Azure | Use storage service encryption customer managed keys for at rest and TLS for in transit |
| Feature flags usage | Toggle new features or model behavior in production without redeploying to test or rollback safely |
| Horizontal scaling vs vertical scaling | Horizontal adds more instances for concurrency vertical increases resources per instance for throughput per request |
| How to implement BYOK | Bring your own key by uploading and managing keys in Key Vault or HSM and configure services to use customer managed keys |
| How to monitor model drift | Track input distribution performance metrics label drift and set alerts when thresholds are exceeded |
| How to restrict model access with AAD | Use Azure Active Directory for authentication protect endpoints with App Registrations and validate tokens on requests |
| How to secure keys and secrets | Store keys in Azure Key Vault enable soft delete purge protection use access policies and RBAC and use managed identities for access |
| How to use Azure Policy to enforce CMK | Create policies that require storage and compute resources to use customer managed keys and audit non compliant resources |
| Incident response for model breaches | Contain affected endpoints revoke credentials analyze root cause remediate data exposure and notify stakeholders |
| Least privilege principle | Grant the minimum permissions required for a role or agent to perform its tasks to reduce attack surface |
| Logging best practice for ML services | Log inputs outputs metadata and decisions avoid logging sensitive raw data use anonymization and use retention policies |
| Managed identity vs service principal in Azure | Managed identities are platform managed identities for Azure resources while service principals are app identities for non managed use cases |
| Mitigations for adversarial examples | Use input preprocessing adversarial training detection mechanisms and model ensembling |
| Model sharding meaning | Splitting a large model across multiple devices or nodes to parallelize inference and reduce memory per node |
| Model versioning best practice | Use immutable model artifacts semantic versioning and metadata to track lineage and enable rollbacks |
| Monitoring SLOs and SLIs for models | Define service level indicators like latency error rate and accuracy then set objectives and monitor adherence |
| Network isolation for multi agent systems | Use VNet peering NSGs subnet segmentation and private endpoints to limit lateral movement |
| Privacy impact assessment purpose | Evaluate risks to personal data for a project and define mitigation controls before deployment |
| Rate limiting and throttling benefits | Protect services from abuse control costs and maintain quality of service under high load |
| Secrets rotation policy | Automatically rotate keys and secrets on a schedule and enforce rapid rotation after suspected compromise |
| Secure container image practices | Use minimal base images sign images scan for vulnerabilities and store in private registries |
| Secure model publishing workflow | Approve models via governance pipeline validate privacy and security artifacts and then deploy with RBAC enforced access |
| Techniques to prevent prompt injection attacks | Input sanitization context separation instruction filtering and model output monitoring |
| Testing strategies for multi agent coordination | Use integration tests fault injection load tests and scenario based evaluations to validate interactions |
| What is Azure Monitor for ML | A platform to collect logs metrics and alerts from models and infrastructure for observability |
| What is Azure Policy for model governance | A service to enforce rules and effects across resources to ensure compliance and governance |
| What is customer managed key CMK advantage | Gives customers control over key lifecycle and import export policies for compliance and separation of duties |
| What is model validation in CI CD | Automated checks including functional tests performance tests bias tests and input output schemas verification |
| What is private link or private endpoint | Provides private network access to Azure PaaS services via private IP removing public internet exposure |
| What is role based access control RBAC in Azure | A model to assign permissions to users groups and service principals using roles at scope levels |
| When to use spot VMs | For fault tolerant stateless workloads and batch jobs to gain large cost savings at the risk of eviction |
About the Flashcards
Flashcards for the Microsoft Multi-Agent AI Solutions Expert exam help students review core Azure security, governance, deployment, and monitoring vocabulary and concepts related to machine learning models and multi-agent systems. They are designed for quick recall of definitions, platform features, and practical controls you'll need to apply under exam conditions.
Deck topics include identity and RBAC (managed identities, service principals), key and secret management (Key Vault, CMK, BYOK, secrets rotation), network isolation and Private Link, Azure Policy and compliance, CI/CD model validation and secure container practices, deployment patterns (canary, blue-green, feature flags), monitoring (Azure Monitor, SLOs/SLIs, model drift), privacy and robustness (differential privacy, data minimization, prompt injection and adversarial mitigations), plus logging, incident response, scaling, and cost optimization.
Topics covered in this flashcard deck:
- Identity and RBAC
- Key management & CMK
- Network isolation & Private Link
- Deployment strategies (canary, blue-green)
- Monitoring, SLOs, model drift
- Privacy and adversarial defenses