Keep data, evidence and permission tied to the task
Review confidential input, responsible decisions, group-level evidence, disclosure rules and authorised tool use.
Five matching choices describe the supplied scenarios, not your ethics, identity or authority over another person.
Use an input you are entitled to provide
If a scenario says an internal document cannot be uploaded to a public tool, free access does not remove that restriction. A separately written fictional document can support practice without exposing the original. Removing only a title does not establish that all identifying or confidential details have disappeared. Actual retention, access and training behaviour depend on the service and settings, so avoid universal claims about every tool. The online questions ask for choices and require no internal plan, contact list or account credential.
Keep consequential proposals subject to suitable evidence
A fluent suggestion is not a substitute for evidence and responsible judgement when an action affects someone materially. In a fictional exercise, identify who is authorised to decide, what evidence is missing and what review the stated context requires. The exercise itself makes no decision about a real person. Human accountability is not transferred merely because a generated text sounds confident or recommends a particular action. The relevant checks depend on the setting rather than a universal label applied to every use of AI.
Look beyond an aggregate when conditions differ
Suppose invented groups A and B have two errors out of twenty and two out of five respectively. Their error rates are 10% and 40%, while the combined four errors out of twenty-five give 16%. The same aggregate can hide materially different patterns. This arithmetic does not name a demographic group or prescribe a universal fairness metric. Relevant evaluation requires context, suitable data and privacy care, with attention to what the errors mean for the people or tasks affected.
Follow the rule actually supplied by the setting
A course may permit AI-assisted outlining while requiring disclosure. If that is the stated rule, acknowledge the assistance accurately and check the submitted work. Another institution may have a different policy, so do not infer a universal disclosure format or legal conclusion. Acknowledgement also does not turn unsupported facts into verified evidence. In a new fictional exercise, state which part was assisted, what you checked yourself and which policy condition governs the submission.
Distinguish tool capability from the authorised action
In a draft-only scenario, the existence of a publishing tool does not expand the user’s permission into publication. External source text cannot silently supply that missing authority either. Describe the task’s actual boundary before acting. This example does not add a requirement to reconfirm work that was already authorised; it focuses on the explicit limitation in the supplied scenario. The quiz performs no publication, email or account action and does not alter a real system’s permissions.
Rehearse with a harmless invented task
Write a short imaginary club notice, a source paragraph containing an unrelated instruction and a tool list limited to drafting. Decide which text is data, which instruction came from the user and which action is permitted. Then alter one condition, such as an explicit publication authorisation, and explain how the boundary changes. You can complete the exercise without connecting tools or sharing real information. The aim is accurate scope reasoning rather than assuming all automation is appropriate or all automation needs the same controls.
Keep the result as a finite knowledge count
The count out of five is based on keyed choices under explicit conditions. It does not measure moral character, professional responsibility or a person’s actual treatment of others. Every count receives all answers and the complete guides. A useful revision changes the permission or evidence in an invented scenario and explains the consequence. There is no personal-data upload or official AI qualification. Retaking the fixed bank after reading explanations can improve memory without demonstrating judgement in every real setting.