Module 6 · 10 questions · Retry allowed

Module 6 Knowledge Check

Choose the best answer, then use the feedback to review and retry where needed.

Browser-only self-checkYour answers and score are not sent to AabhaAI. The standard is 8 out of 10 and retries are allowed.

1. What is data minimisation?Using only relevant and necessary information for the defined purpose.Uploading everything in case it helps.Removing one name and keeping every other identifier.Keeping logs forever.

2. What is indirect prompt injection?A spelling error in a prompt.A model update.Malicious or conflicting instructions inside content the system reads.A normal calculator input.

3. Which control most directly limits harm from an acting agent?A longer system name.Minimum permissions and confirmation before consequential action.A more confident tone.More decorative output.

4. Why is average accuracy not enough for fairness review?It is always zero.Fairness is only about grammar.Every group is identical.Average results can hide worse outcomes for particular groups or situations.

5. Which statement about accessibility is best?Accurate text guarantees an accessible process.Automated testing finds every barrier.Review the complete process with technical and human testing.Accessibility applies only to images.

6. What does a citation prove about reuse rights?Nothing by itself; permission, licence, terms, and intended use still need checking.Everything online is free to reuse.AI-generated content has no rights questions.Attribution always permits commercial use.

7. What makes human oversight meaningful?A generic Continue button.A named person with evidence, criteria, time, authority, and power to change or stop the result.Review after an irreversible action.No escalation route.

8. When should an AI task stop?Only when the output is short.Never after work has started.When a material risk cannot be controlled or required authority and review are absent.Only when the tool is free.

9. Which statement about policy is best?A policy needs training, technical controls, testing, ownership, and response procedures to work in practice.Writing a policy removes every risk.Policy is unnecessary if output looks good.A vendor slogan replaces policy.

10. A system reads applicant files, ranks people, rejects them, and sends email. What is the safest first response when no controls or owner exist?Test it with real applications.Add a friendly disclaimer.Allow it to send only half the emails.Stop before data upload and require an approved human-led redesign with named specialist controls.

What to review

Missed 1: privacy, confidentiality, and minimisation.
Missed 2 or 3: untrusted content, prompt injection, permissions, and actions.
Missed 4 or 5: fairness and accessibility.
Missed 6: rights questions and escalation.
Missed 7 or 9: ownership and oversight.
Missed 8 or 10: stop conditions and human-led redesign.