Lab: Policy draft
Draft a one-page, actually-readable AI usage policy for your team — using AI, correctly, to do it.
Most corporate AI policies are either 'anything goes' silence or ten pages of legal fog nobody reads. In this lab you'll draft the useful kind — one page, concrete, written for your actual team — and you'll use AI to draft it, which is itself a demonstration of everything in this module.
Step 1 — Collect the raw facts (you, 5 minutes)
- 1Which AI tools does your team actually use today, officially or not?
- 2Which of the four data classes does your team handle?
- 3The three most common AI tasks on your team (from your Module 3 pattern-spotting).
- 4Any rules that already exist (IT policy, client contracts, regulator guidance).
Step 2 — Generate the draft
You are drafting a one-page AI usage policy for a real team. Team context: [your step-1 facts]. Requirements: written in plain language a busy person reads in 3 minutes; organized as Green (encouraged — with our 3 common use cases as examples), Yellow (allowed with named safeguards: redaction, verification, disclosure), Red (never — with our data classes named specifically). Include a 2-line 'why this exists' intro that isn't scary, and a 'when unsure, ask [ROLE]' escalation line. No legal boilerplate.
Step 3 — Red-team your own policy
Now attack this policy. You are (a) a well-meaning employee looking for loopholes to save time, and (b) a cautious lawyer looking for what's missing. List the top 5 gaps or ambiguities each persona finds, then revise the policy to close the ones that matter without doubling its length.
Red-teaming a draft with the same tool that wrote it is a pattern you'll reuse constantly — the model critiques its own work far better than it first-drafts.
Step 4 — The human pass
Read it as yourself. Does the Green section actually cover what your team does daily? Would you follow the Yellow safeguards, or are they aspirational fiction? Cut anything performative. Then — ideally — show it to one teammate and your manager. This one-pager is designed to be adopted as a team's real policy, and bringing a usable draft to your manager is a genuinely good look.
Problem set 4
Six borderline scenarios. For each, rule: allowed, allowed-with-safeguards (name the safeguards), or no — citing which lesson's principle decides it.
- A recruiter pastes forty résumés and asks for a ranked shortlist of the top five.
- A nurse summarizes shift-handover notes, with patient names removed, on a personal account.
- A consultant pastes a client deck covered by an NDA to tighten the executive summary.
- A manager drafts interview questions from the published job description.
- An analyst pastes real customer names and purchase histories to brainstorm segmentation ideas.
- An HR partner drafts the company-wide announcement of an already-decided restructuring from approved facts.