AISG practice · Module 4

Data Governance, Privacy & Protection for AI — 25 questions. Choose the BEST answer.

You can answer as a guest, but you'll need a free account to see your score and save it. Sign up · Sign in
Question 1 Question 1 · M4
A developer pasted a model API key into a system prompt so the agent can call a service. What is the BEST fix?
Question 2 Question 2 · M4
A licensed dataset is found to contain unlawfully collected personal data. What makes a targeted response possible?
Question 3 Question 3 · M4
Which role approves the use of a dataset for a new AI training purpose?
Question 4 Question 4 · M4
An appeals channel lets any user submit 'should be allowed' examples that feed retraining. What is the MOST important control?
Question 5 Question 5 · M4
Which control BEST reduces the risk of a generative model reproducing copyrighted training content verbatim?
Question 6 Question 6 · M4
Which control supports chain of custody for datasets?
Question 7 Question 7 · M4
After a backdoor is discovered, which capability MOST speeds containment?
Question 8 Question 8 · M4
Which evidence BEST shows that a training set was not changed after approval?
Question 9 Question 9 · M4
A dataset has names and IDs removed but keeps postcode, birth date and gender. Why is it NOT safely anonymous?
Question 10 Question 10 · M4
A consent form says 'we may use your data to improve our services and AI'. What is the MAIN weakness?
Question 11 Question 11 · M4
Export-controlled engineering documents are indexed into a vector store accessible to staff in many countries. What is the PRIMARY concern?
Question 12 Question 12 · M4
An organisation wants to train on public social-media posts. What is the BEST position?
Question 13 Question 13 · M4
Why is cryptographic erasure often preferred for sanitising AI data in cloud storage?
Question 14 Question 14 · M4
Which protection addresses data and model weights while they are being processed in memory?
Question 15 Question 15 · M4
Which attack hides behaviour that is activated only by a specific trigger?
Question 16 Question 16 · M4
What is purpose-based access control?
Question 17 Question 17 · M4
Which architecture choice makes honouring erasure requests EASIEST?
Question 18 Question 18 · M4
Which technique helps find training records with outsized influence on a model's behaviour?
Question 19 Question 19 · M4
Which risk is specific to recursive training on model-generated data?
Question 20 Question 20 · M4
A user's data is deleted from source systems, but a nightly-rebuilt vector index and a fine-tuned model both used it. What is TRUE?
Question 21 Question 21 · M4
Which AI pipeline activity is MOST likely to create an unnoticed cross-border data transfer?
Question 22 Question 22 · M4
Which regulation MOST directly governs card numbers appearing in AI chat transcripts?
Question 23 Question 23 · M4
At which lifecycle stage should anomaly and poisoning screening primarily occur?
Question 24 Question 24 · M4
What distinguishes lineage from provenance?
Question 25 Question 25 · M4
What classification should embeddings generated from confidential documents normally carry?