Page 56 - CyberSecurityEssentialsEnglish

↻ Exercise 5: Error analysis
Goal: Diagnose a failure before attempting to fix it.
Required: One weak output you have kept, with the request that produced it. Allow
about 20 minutes.
Steps: State exactly what was wrong. Then test five candidate causes in turn: missing
context, the wrong task type for the tool, outdated information, calculation, and a

change in the data or the conditions of use. Record the evidence for the cause you settle
on.
Output: A short diagnosis naming one cause and the evidence for it.
Assessment criterion: A cause is chosen with evidence. Rewriting the request is not
treated as the answer to every kind of failure.

↻ Exercise 6: Design a collaboration
Goal: Place the human decision explicitly inside the work.

Required: One task shared between you and a system. Allow about 25 minutes.
Steps: Draw the steps in order. Mark for each step whether a person or the system
performs it. Then name who defines the goal, who reviews the output, who approves it,
and where the work stops if something looks wrong.
Output: A step diagram with the three roles and the stopping point marked.
Assessment criterion: The diagram contains a point at which a person can stop the work,

not only a point at which they approve it.

↻ Exercise 7: Evaluate a low-risk use
Goal: Turn a general idea for using AI into a small measurable experiment.
Required: A recurring task that contains no sensitive data, such as summarizing an
internal report or organizing meeting notes. Allow about 60 minutes.
Steps: Define the output type. State the role of the system and the role of the person.
Prepare five local test cases including one incomplete case and one edge case. Define the

acceptance criterion before testing. Then compare the time, the quality, and the number
of corrections before and after use.
Output: One page describing the task, the output type, the role of each side, the test
cases, and a decision to continue or revise the use.
Assessment criterion: Success is not the appearance of the output. It depends on a
defined task, representative cases, a criterion set in advance, and an explicit statement of

what stays under human review.

55