AI in Fraud Detection Quiz

Answers and explanations appear as you take the quiz.

1. What is the main limitation of pure rule-based fraud systems that AI addresses?
   • Criminals learn the fixed rules and route around them
   • They are too expensive to run
   • They cannot process numbers
   • They only work on cash transactions
2. Why are false positives a serious concern in fraud detection?
   • They slow down the servers
   • They legitimately block real customers and erode trust
   • They cost nothing to fix
   • They only affect fraudsters
3. Which technique is especially useful for exposing rings of colluding fraudsters?
   • Image classification
   • Speech recognition
   • Graph analysis of shared devices, emails, and IPs
   • Spell checking
4. What limitation can message-passing GNNs face in imbalanced fraud networks?
   • They guarantee that every linked transaction is fraud
   • They remove the need to monitor changing data
   • They automatically retrain with verified labels
   • Neighbor and hub effects can amplify the dominance of legitimate transactions
5. Should a fraud model be retrained on a fixed frequent schedule regardless of evidence?
   • No; monitor drift and validate retraining against current data and performance
   • Yes; retraining guarantees fraud patterns will be detected
   • Yes; every model becomes invalid after one day
   • No; changing tactics mean monitoring is unnecessary
6. What key term does FinCEN ask institutions to include when filing a SAR related to its deepfake alert?
   • FIN-2024-VOICEPRINT
   • FIN-2024-DEEPFAKEFRAUD
   • FIN-2024-ALL-TRANSACTIONS
   • FIN-2024-IDENTITY-GUARANTEE
7. What does a velocity feature usually measure?
   • The customer’s credit score only
   • The graph’s number of layers
   • The frequency or amount of activity within a specified time window
   • A model’s parameter count
8. Why should AI fraud alerts be reviewed by investigators?
   • A model score is a court judgment
   • Investigators cannot see transaction data
   • AI prevents all false positives
   • Model outputs may be wrong or incomplete and do not establish legal culpability