212-82 Question #135: Real Exam Question with Answer & Explanation

Question

As the director of cybersecurity for a prominent financial Institution, you oversee the security protocols for a vast array of digital operations. The institution recently transitioned to a new core banking platform that integrates an artificial intelligence (Al)-based fraud detection system. This system monitors real-time transactions, leveraging pattern recognition and behavioral analytics. A week post-transition, you are alerted to abnormal behavior patterns in the Al system. On closer examination, the system is mistakenly flagging genuine transactions as fraudulent, causing a surge in false positives. This not only disrupts the customers' banking experience but also strains the manual review team. Preliminary investigations suggest subtle data poisoning attacks aiming to compromise the Al's training data, skewing its decision-making ability. To safeguard the Al- based fraud detection system and maintain the integrity of your financial data, which of the following steps should be your primary focus?

Options

• ACollaborate with the Al development team to retrain the model using only verified transaction data
• BMigrate back to the legacy banking platform until the new system is thoroughly vetted and all
• CLiaise with third-party cybersecurity firms to conduct an exhaustive penetration test on the entire
• DEngage in extensive customer outreach programs, urging them to report any discrepancies in their

Explanation

To address an AI system mistakenly flagging genuine transactions as fraudulent due to data poisoning, the most effective action is to retrain the model with verified data and implement robust data integrity checks.

Common mistakes.

• B. Migrating back to the legacy banking platform is an extreme and disruptive measure that avoids addressing the technical issue with the new system and negates the investment in the advanced fraud detection capabilities.
• C. Liaising with third-party cybersecurity firms for a penetration test primarily identifies vulnerabilities in the system's defenses, but it does not directly fix the current state of a compromised AI model that requires retraining with clean data.
• D. Engaging in extensive customer outreach is important for communication and managing customer impact, but it does not resolve the underlying technical problem of the AI model's misclassification caused by data poisoning.

Concept tested. AI security, data poisoning remediation

Reference. https://www.microsoft.com/en-us/security/business/security-101/what-is-ai-security

No community discussion yet for this question.