Certified AI Safety & Security - Practice Quiz — Quiz in Progress

Q1. Which best captures the key idea of "Prompt injection basics"? ATrack every change, undo mistakes, and collaborate without overwriting each other BAdd rules to keep outputs appropriate for your audience CRestate the persona if a long chat starts drifting from it DPrompt injection is when untrusted text tries to override your instructions

Q2. Which best captures the key idea of "Defending against injection"? AGive data in a clean structure (tables, JSON) so the model parses it reliably BCompare planned vs actual weekly; surface slips early CSeveral specialised agents collaborate, each owning a role DSeparate instructions from data, and tell the model to treat data as untrusted

Q3. Which best captures the key idea of "Avoiding data leakage"? AWrite clear briefs for a campaign BDon't put secrets in prompts; assume anything sent could be logged CUse AI to spot missing values, outliers, and formatting issues DPersist key facts across sessions by saving and re-injecting a profile summary

Q4. Which best captures the key idea of "Refusals and boundaries"? ATurn ranges into bar, line or pie charts; pick the type that matches the... BAnalogies, odd-one-out and comprehension sharpen language logic CUse AI to draft, explain, and review code while you stay in control DDesign prompts that decline unsafe or out-of-scope requests gracefully

Q5. Which best captures the key idea of "Reducing bias"? ADiverge widely first (quantity), converge later (criteria) BAwareness → interest → decision → action; fix the leakiest stage first CMatch hard work to peak hours; sleep, movement and breaks are performance tools DAsk for balanced perspectives and check outputs for stereotyping

Q6. Which best captures the key idea of "Content safety"? ACreate a test set and score your workflow's outputs BDeepSeek is strong at code and reasoning; give precise specs CCreate questions with answer keys and difficulty DAdd rules to keep outputs appropriate for your audience

Q7. Which best captures the key idea of "PII handling"? ARedact personal data and instruct the model not to infer or store it BData lives in tables of rows and columns, defined by types CGood businesses solve painful problems for people who can pay — validate that first DRecognise attempts to override instructions

Q8. Which best captures the key idea of "Responsible disclosure of limits"? ATell users when AI is uncertain or when a human should decide BWrite clear instructions so others can reuse your prompts CPersonas built from real data guide every message and channel choice DCondense long threads into the key issue and status