Data Security and Privacy; Continuous Learning and Adaptation
Free CSA Trusted AI Safety Expert (TAISE) practice — 6 questions on Data Security and Privacy; Continuous Learning and Adaptation, with explanations. No sign-up. Full 12-question mixed test →
A healthcare AI company trains a diagnostic model using patient records held across five hospitals. Regulatory and institutional policies forbid centralizing the raw patient data, but the team wants to continue improving model accuracy collaboratively. Which approach BEST meets these requirements?
Federated learning with secure aggregation allows each hospital to train locally and share only encrypted/aggregated model updates, meeting the no-raw-data-sharing constraint while still improving the shared model — the standard pattern for this exact cross-institutional scenario.
A support-ticket dataset used for LLM fine-tuning contains free-text fields with embedded names, emails, and SSNs. The team must comply with data minimization principles in the training-data governance program while still fine-tuning for response quality. Which approach is correct?
Data minimization requires removing or pseudonymizing unnecessary PII before data enters the training pipeline, retaining only fields needed for the task — preventing the model from memorizing and later leaking sensitive identifiers.
Which statement correctly distinguishes concept drift from data drift in AI continuous monitoring?
Concept drift is a change in the mapping between inputs and the target (the 'concept' the model is learning), whereas data drift is a shift in input feature distributions without a change in that underlying relationship — a key distinction TAISE tests for continuous monitoring design.
During a training-data governance review, a dataset is proposed to satisfy k-anonymity with k=2 on its quasi-identifiers before external sharing. What is the MAIN concern a reviewer should raise?
k=2 means a record can be narrowed down to a group of just two, making re-identification via linkage attacks (e.g., combining with external datasets) comparatively easy; higher k values are typically required for meaningful privacy protection.
A deployed LLM API begins receiving a high volume of systematic queries following a grid-search-like pattern across the input space, and shortly afterward a competitor's product shows suspiciously similar output behavior. Per TAISE guidance on continuous monitoring and incident response, what should the team do FIRST?
The query pattern is a classic model-extraction/model-stealing indicator; the appropriate first response is to detect, contain (rate-limit), and investigate per the incident response plan before deciding on further escalation — balancing availability with security.
An AI governance team wants its safety and compliance program to remain effective as regulations such as the EU AI Act evolve and new threat intelligence emerges. Which practice BEST satisfies the TAISE principle of continuous learning and adaptation?
Continuous learning and adaptation requires an ongoing process — horizon-scanning, periodic reassessment, control updates, and staff training — rather than a static, one-time activity, to keep pace with evolving regulation and threat landscapes.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →