Time: about 12 minutes. Read the unit page first.
Part 1 — answer the three questions about your own tool (4 minutes)
Pick the AI service you actually use most. Find out, from its own documentation or settings:
| Question | What you found | Where you found it |
|---|---|---|
| Is my input used to train future models? Can I change that? | ||
| How long are conversations kept, and who can read them? | ||
| Where is it processed? |
Also note which of saved memory, shared links and connectors are on for you, and the date you looked. Asking the assistant itself does not count: its answer is generated text.
If you have no AI access, or use no AI service, pick one your institution offers or one you have heard of, and answer from its published policy. You need no account; for "can I change that", record what the policy says the default is.
Two rules for this part.
Record where you found each answer. This is the discipline Unit 9 applies to citations, used here on your own tools. A locator makes the answer checkable and dateable; without one you have a memory.
"I could not find out" is an answer. Write it, and note how long you looked. If a provider does not make this findable in a few minutes, that is a fact about the service and it should push your classification stricter.
Part 2 — sort eight items (4 minutes)
Put each into one band, or say why it sits in two: public, yours alone, someone else's, not yours to decide.
| Item | |
|---|---|
| 1 | A paragraph from a published journal article |
| 2 | Your own half-finished essay |
| 3 | A screenshot of a group chat, to ask what someone meant |
| 4 | An internal spreadsheet of your employer's monthly figures |
| 5 | A CV a friend sent you, to ask how to improve it |
| 6 | Your own lecture notes, typed by you in a lecture |
| 7 | An anonymised case study from a textbook |
| 8 | A draft paper a colleague shared for comment |
Then answer one question in a sentence: which one did you find hardest, and why? There is a defensible disagreement available on at least two of them, and noticing where it is matters more than the label.
Part 3 — redact one, and test it (4 minutes)
Take item 5, or any real piece of someone else's material you might genuinely have wanted help with.
- Rewrite it so the task survives and the person does not appear: remove names, dates, identifiers, and distinctive details.
- Then apply the re-identification test: could a colleague who knows the setting work out who this is? Look particularly at combinations — a role plus a place plus a time is often enough in a small institution.
- If the answer is yes, generalise further or abandon the task.
Write down what you removed and what you had to change beyond names. The second list is usually the interesting one.
Now take the checkpoints
cp13-what-leaves is about the reading's section on what leaves the device.
cp13-whose-material is about Part 2.
python3 selfcheck.py run cp13-what-leaves # or in the browser
python3 selfcheck.py run cp13-whose-material # or in the browser
Run the Python route from The verification lab and self-check, or use the browser self-check.
Optional extensions
Optional. These sit outside the core study time, are not required for the checkpoints, and nothing later in the course depends on them.
Find your institution's actual rule (about 6 minutes). Locate the policy that governs AI use with personal or institutional data where you study or work, and record where it is and what it says about the four bands. This course cannot tell you the rule; it can tell you that you should know where it lives. Note whether it addresses the "not yours to decide" band at all — many do not, and knowing that is useful.
Open an attachment as text (about 4 minutes). Take a document you might have attached and look at what a text extraction of it actually contains: tracked changes, comments, metadata, hidden rows, the second worksheet. The gap between what you thought you were sending and what would have gone is the point of the exercise.
Marking guidance — open this once you have done the activity, to check your own work
Read this after attempting the activity.
Part 1 — what a good answer looks like
The locator column is what separates this from a vague impression. "The privacy policy, section on data retention, read today" is an answer; "I think it does not train on my data" is a memory, and this is the course that distinguishes those.
If you found the answers in a marketing page rather than a policy or a settings screen, note that. Marketing text about privacy is a claim by someone who gains from it, which is a Unit 9 problem. To judge whether a source deserves to be relied on, leave it and see what others say about it (Unit 9, where your route includes it, calls this reading laterally). Cite the policy itself.
"I could not find out" pushes the classification stricter, not looser. An unanswerable question about retention means you treat the service as if the answer were the least favourable one, because you have no basis for anything else. Unit 12 makes the same move for a risk you cannot assess.
Part 2 — the eight items
| Band | Why | |
|---|---|---|
| 1 | Public | Already published. |
| 2 | Yours alone (for privacy) | See the reading on the separate assessment question — that is not a privacy matter and has a different authority. |
| 3 | Someone else's | Other people wrote those messages, in a setting they took to be private. Your being in the chat does not make it yours to publish. |
| 4 | Not yours to decide | The dullness of the figures is irrelevant. It is the employer's material and probably under an agreement someone else signed. |
| 5 | Someone else's | A CV is dense with personal data — name, employers, dates, sometimes address. Sending it to help is still sending it. |
| 6 | Yours alone, with a caveat | You typed them, so the words are yours. If they quote a named person's unpublished remarks, that part is not. |
| 7 | Public | Published and already anonymised by someone with the responsibility to do so. |
| 8 | Not yours to decide, and someone else's | Unpublished research shared in confidence. Two bands at once, and both say ask first. |
The two where reasonable people differ are 6 and 7.
Item 6 divides on whether lecture notes are your words or a record of someone else's. In practice both: your paraphrase is yours, a verbatim quotation of an unpublished remark is not, and the useful habit is noticing that a single document can span bands.
Item 7 divides on how much you trust the anonymisation. Published and professionally anonymised is a reasonable basis; "it looked anonymised to me" is not, and the re-identification question from Part 3 applies to material you receive as well as material you produce.
Item 4 is the one most people get wrong, and it is worth dwelling on. Nothing about a routine internal spreadsheet feels sensitive, which is exactly why the band exists: the decision is not yours to make on the material's apparent importance, because you are not the person who agreed the terms.
Part 3 — what you had to change beyond names
The second list is where the learning is. Common answers:
- a job title that only one person holds;
- a date that pins the event;
- an unusual circumstance that is the whole reason you were asking;
- the name of a small organisation, which narrows the population to a handful.
That last category is the awkward one, because the distinctive detail is often the thing you needed help with. When removing it destroys the question, you have found a task that should not be sent — and that is a result, not a failure of redaction.
If you concluded that no colleague could identify the person, check whether you tested that against someone who knows the setting or against yourself. You are the worst possible judge of whether your own redaction works, because you already know the answer.
The checkpoints
cp13-what-leaves tests the model of what a hosted service receives. The trap is
thinking in terms of what you asked about rather than what you sent.
cp13-whose-material tests the classification. The trap is sorting by how
sensitive material feels rather than by whose it is — which is why the internal
spreadsheet is in the question.
Your private activity record
Browser storage is not a permanent copy
Progress is kept only in this browser, profile and device. Private browsing, clearing site data, removing the profile, a browser reset, storage eviction or device loss can erase it. Keep important answers and contributions separately.
These notes stay in this browser unless you download a backup or activity log.