CAIC Cheat Sheet 2026

The 30 highest-yield CAIC facts, distilled from real exam questions. Print it, save it as a PDF, or study it here — free, no sign-up.

1. Which mechanism allows an AI API server to inform clients that it should only be accessed over HTTPS for a specified duration? → HTTP Strict Transport Security (HSTS)
2. A CAIC consultant recommends separating AI training and inference workloads into different node pools. What is the PRIMARY reason? → Training and inference require different hardware profiles and scaling behaviors
3. An AI consultant is asked to evaluate model fairness. Which metric detects whether a model performs significantly worse for a protected demographic group? → Disparate impact ratio
4. Which approach enables an LLM to use real-time internet data during inference without retraining the model? → Web-augmented retrieval (search tool integration)
5. Which database type is most architecturally suited as a vector store for similarity search in a RAG (Retrieval-Augmented Generation) system? → Purpose-built vector database with ANN index support like Pinecone or Weaviate
6. The EU AI Act classifies AI systems into risk tiers. Which category includes biometric surveillance in public spaces? → Unacceptable risk
7. Which storage class is most cost-effective for storing AI training data that is accessed once per month for retraining? → Infrequent access (warm) storage
8. What is an SLA in CAIC technology? → A formal agreement defining expected service performance standards
9. Which approach to AI governance uses a tiered, risk-proportionate regulatory framework? → Risk-based regulation (as in the EU AI Act)
10. In the context of AI SDLC, 'concept drift' requires which operational response? → Monitoring model performance metrics and triggering retraining
11. A generative AI model produces realistic fake images of real individuals. The primary ethical concern is: → Non-consensual synthetic media enabling defamation and identity harm (deepfakes)
12. In AI supply chain security, which risk is introduced when using open-source model repositories without verification? → Malicious actors may publish tampered models with embedded backdoors or trojans
13. A company trains an AI model using customer purchase data without explicit consent. Which compliance violation is most directly applicable? → Violation of data privacy regulations such as CCPA or GDPR
14. What is the security risk of using wildcard TLS certificates (*.example.com) for AI microservice endpoints? → Compromise of one service's private key exposes all subdomains
15. A team wants to ensure their AI model can be rolled back quickly in production. The BEST DevOps practice to support this is: → Blue-green deployment with model registry versioning
16. What is 'hallucination' in the context of large language models? → Model generating confident but factually incorrect outputs
17. When should an AI consultant recommend using serverless inference (e.g., AWS Lambda, Google Cloud Run) instead of a dedicated GPU instance? → For lightweight, infrequent inference tasks where cold starts are acceptable
18. A cloud AI platform uses object storage for model artifacts. Which versioning strategy ensures traceability between code, data, and model? → Use content-addressable storage with SHA-256 hashes tied to git commit IDs
19. What is the principle of least privilege in CAIC technology? → Users get only the minimum access needed for their role
20. Which cross-validation strategy is most appropriate when the dataset has a temporal ordering and future data must not leak into training? → Time series split (walk-forward validation)
21. What is machine learning in the context of AI? → Systems that learn and improve from data
22. In the context of neural networks, what does 'vanishing gradient' refer to? → Gradients becoming extremely small in early layers, stalling learning
23. A company uses AI to infer employee mental health status from productivity data without disclosure. This violates which ethical principle? → Non-maleficence and privacy
24. A dataset has 95% negative samples and 5% positive samples. Which metric is LEAST informative for evaluating a classifier on this dataset? → Accuracy
25. Which of the following is a key principle of 'privacy by design' as applied to AI systems? → Embedding privacy controls proactively into system architecture from the start
26. The concept of 'differential privacy' in AI systems primarily protects against: → Membership inference attacks that reveal individual training records
27. What is scope creep in CAIC project management? → Uncontrolled scope expansion without adjusting time, cost, or resources
28. A team is conducting a 'post-mortem' after an AI model caused a significant error in production. What SDLC artifact is MOST useful as input to this review? → Model monitoring logs and prediction audit trail
29. Why is AI scalability important for businesses? → It supports efficient growth and adapts to increasing demands
30. Which cloud AI service allows consultants to deploy, version, and monitor custom ML models without managing underlying infrastructure? → AWS SageMaker