Lesson 8 of 8 · 9 min read · last verified 2026-08-26
What not to hand over
In this lesson you will:
- Apply a confidentiality test before pasting anything into an assistant
- Identify work whose difficulty is the point and should stay yours
Every previous lesson expanded what you hand over. This one draws the line, and it is the most important lesson in the module — a productivity routine without a boundary is just a faster way to make a serious mistake.
Three categories. They are not equally obvious.
What you paste
The most common workplace incident with AI is not a dramatic one. It is somebody pasting something they should not have into a box, because it was the fastest way to get a task done.
Use the paste test:
Would I be comfortable if this exact text appeared in an email to someone outside my organisation?
If not, it does not go in. That covers customer records, unreleased financials, anything under NDA, colleagues’ personal details, medical or immigration information, credentials, and internal documents about people.
Two things people get wrong here:
“It’s just a summary.” The summary contains the confidential part, or you would not need it summarised.
“Enterprise means private.” Enterprise tiers usually mean your text is not used for training. Not that it is never stored, never processed elsewhere, or covered by whatever obligation you personally are under. Check your organisation’s actual policy — and if there isn’t one, that is not permission.
Where the material is sensitive but the task is real, change the input: remove the names, replace the figures with plausible fakes, or describe the shape of the problem rather than pasting the thing.
Decisions about people
Hiring, firing, promotion, discipline, grading, credit, benefit, suspicion.
Not because a model produces obviously bad output on these — the output often reads well, which is the danger. Two reasons:
Accountability. Somebody has to be answerable to the person affected. A tool cannot be, so the responsibility does not actually move — it just becomes easier to feel it has.
The person is owed a human judgement. Being sorted by something that cannot be argued with is a specific kind of harm, distinct from being sorted wrongly.
Assistants can still help around the edges: checking your written reasoning is clear, or asking what you might have missed. The judgement stays with you, and so does the conversation.
When the difficulty is the point
The subtlest category, and the one that costs you slowly.
Some work has effort as the product. Thinking through a hard problem is how you come to understand it. Writing an argument is how you find out whether you believe it. Struggling with a subject is how it gets learned — including, quite directly, this course.
Hand those over and you get the artefact without the thing the artefact was for. The document exists. The understanding does not. That is invisible in the moment and expensive later, in the meeting where you have to defend it.
A test that works: will I need to defend, extend, or build on this? If yes, do the thinking. Use the assistant to check it afterwards, which is a genuinely good use — one that strengthens your understanding rather than substituting for it.
Try it now (7 minutes)
Look back at the routine you built in L7. For each step, ask the paste test. If any step involves material that would fail it, rewrite that step to remove the names and figures — or drop it.
Then ask whether any step is doing thinking you will later have to defend.
Check your understanding
Recap
Apply the paste test before anything goes in, keep decisions about people with the human who must answer for them, and do the thinking you will later have to defend. The boundary is what makes the rest of the module safe — and it completes E4: you now know where AI fits in your day, how to use it for email, reading, research, notes and planning, and where to stop.
🗂 3 flashcards from this lesson join your daily review.
Previous: Your twenty-minute routine