Deployment Labs
Module 2 · Lesson 7 of 7 · 8 min

Your starter guardrails

All six guardrails on one page, ready to publish before day 0.

This is everything from the module on one page: the approved tools, the data map, the output standard, the spending limits, stop, tell, fix, learn with its contacts, and the rules for agents. Your sponsor signs it off, IT or compliance keeps it current, and you publish it where people already work, before day 0.

Anything still missing is flagged below, with a link back to the lesson where you add it.

Starter guardrails · The Deployment Framework

Your organisation

Your first 90 days with AI

The guardrails
Six guardrails that make the safe way the easy way, set up before anyone is asked to change how they work.
1 · Approved tools

Tools that aren't on the list aren't banned in secret. Raise them at the front door. Every new tool answers the eight vendor questions before it's approved: training on our data, storage and deletion, the agreements we need, sign-on and removal, admin control of AI features, subprocessors and models, the way out, and full cost with a hard cap.

2 · Data map
DataExamplesAllowed in
PublicWebsite copy, published pricesAny approved tool
InternalPlans, internal emails, non-personal numbersApproved tools on business plans
ConfidentialClient work, contracts, personal data about staffOnly tools marked “confidential OK”
RegulatedPatient data, card data, anything under HIPAA or similarOnly tools with the required agreement in place (a BAA for HIPAA)

When something could fit two classes, treat it as the higher one.

3 · Output standard
  1. 1Labelled: Say where AI helped.
  2. 2Owned: Your name is on it, and you can explain it.
  3. 3Short: No longer than the reader needs.
  4. 4Checked: You read it and checked the facts that matter before sending.
Label line
Not set yet. Add it
4 · Spending limits
Ceiling (aim page)
Not set yet. Add it
Monthly cap, all tools
Not set yet. Add it
80% alert goes to
The person who runs the loop, and finance

Usage is never a performance measure. Spending is reported per tool, never per person.

5 · Stop, tell, fix, learn
  1. 1Stop. Stop it, if you can do so safely.
  2. 2Tell. Within one working day, tell not named yet, or not named yet if it's sensitive.
  3. 3Fix. The owner contains it and puts it right.
  4. 4Learn. A short blameless note goes on the friction list.

Nobody is blamed for reporting.

6 · Rules for agents
  • Its own login, with limited accessEvery agent signs in as itself, never as a person, and can only reach what its job needs.
  • Reach is set in the toolsWhat it can reach is enforced by permissions in the systems, not just written in its instructions.
  • It stops itself, and anyone can stop itIt halts when its output looks wrong, and anyone who sees a problem can press stop.
  • Reach is earnedNew builds can't reach live systems until they're promoted to a limited live trial, and only get full reach once they're kept.
First agent
Not set yet. Add it

The pilot kit's minimum set also includes a first AI list: every tool and use already running, from three sources. You start it in module 3.

Signed off by the sponsor
Maintained by (IT, security or compliance)
Date
Your output

Get the print-ready starter guardrails page

Your details unlock printing, saving as PDF and copying the starter guardrails page. A copy comes to Deployment Labs so we can help if you want it.

  • Print or save as PDF, ready to publish
  • Copy as text for your intranet or handbook
  • The before-day-0 checklist from the pilot kit

Taking the course with your leadership team? Use the same team code, and your answers can be compared when we talk.

By continuing, you agree to receive emails from Deployment Labs about this course and related resources. You can unsubscribe at any time. Privacy policy

What happens next

The guardrails make the safe way the easy way. Module 3 puts them to work. Under the promises from module 1, you invite everyone to say what AI they already use, check that against your sign-on and expense records and the software that has switched AI on, and start your AI list. Most of what turns up will land on the approved tools list you just built, or go through the front door to get there.

The guardrails get a full review once a year, and the loop refines them as you go: every blameless note on the friction list is a chance to make one of them easier to follow.

If you'd like help running it, Deployment Labs takes leadership teams through this first loop as a guided pilot. Book a call and bring your aim page and your starter guardrails.

Finished this lesson?

Mark it complete. Your answers are already saved in this browser.