Co-written AI policy

For NGOs, nonprofits and foundations: an artificial intelligence use policy people apply, not a charter people sign. Your team writes it in three workshops; we hold the pen and the method.

Length
Four to six weeks
Method
Three workshops with a pilot team, a test on ten real cases
Pilot team
5 to 8 people from your organisation, field staff included

How is it different from a charter?

A charter states principles. A policy says what to do, on Monday morning, faced with a specific case. Its amber light is wide, its red short and justified: a rule that is too strict pushes use into the shadows.

What does the policy include?

Two levels of text, and the toolkit to apply them.

  1. The charter
    One to two pages: the principles and the red lines. What you display, including to your donors.

  2. The use policy
    Six to ten pages: scope, uses in green, amber or red, data rules by sensitivity and account type, human checking, transparency, images, buying a tool, training, incidents, review.

  3. The register of tools and uses
    For each tool: the subscription, the data allowed, the person in charge, the review date.

  4. The decision tree
    One page: “Can I use AI for this?” The policy in thirty seconds.

  5. The procedure to approve a new use
    A one-page form, with the threshold that triggers an impact assessment. So that no department decides alone.

  6. The tool purchasing grid
    The clauses to require from a supplier: audit right, notice of model changes, data ownership, exit right.

  7. The transparency statement
    Standard wording for your reports, your publications and your images.

  8. The incident reporting form
    What happened, who to tell, what to do. To learn from mistakes.

  9. The integration map
    What changes in each of your policies: data protection, safeguarding, code of conduct, IT charter, communication, procurement, human resources.

  10. The roll-out kit
    A one-page sheet for all staff, ten questions and answers, a fifteen-minute presentation for team meetings.

How do we write the policy?

Over four to six weeks, in the full version.

  1. Preparation

    Before the workshops

    We read your existing policies and, if there was one, your diagnostic.

  2. Workshop 1

    3 h

    Principles and red lines; your real uses sorted into green, amber or red; your data sorted by sensitivity.

  3. First version

    Drafting

    We adapt it to your law and to sector standards, and integrate it into your existing policies.

  4. Workshop 2

    3 h

    The edge-case test: ten real cases brought by the team, such as translating a victim’s testimony or illustrating a publication with a generated image. The policy must settle each one clearly, or we fix it.

  5. Second version

    Drafting

    The corrected policy, and its annexes.

  6. Workshop 3

    2 to 3 h

    Governance and roll-out: roles, the register filled in during the session, the communication plan, indicators, the review date.

  7. Adoption

    1 h

    A presentation to your leadership or your board. In France, informing or consulting the works council (CSE) where required.

  8. 3-month check

    In the full version

    A short survey, a check of the register, and the edge-case test run again.

What we ask of you

  • A pilot team of 5 to 8 people: leadership, programmes, monitoring and evaluation, safeguarding, communication, IT or administration, human resources, and ideally a board member.
  • Field staff in contact with communities, in the pilot team or consulted. A wider consultation of communities is possible as an option.
  • Your existing policies.
  • Ten real cases for the edge-case test.
  • Two conditions written into the quote: your leadership present at workshops 1 and 3, and an adoption date set from the start. Without them, a policy stays a dead letter.

What do you build on?

Every legal point is written as a point to be checked by a lawyer or your data protection officer.

How do you know it is applied?

We test a policy the way we test an evaluation instrument: by putting cases to it.

  1. Share of staff who know the policy

    Reading or training certificate

  2. Share of tools actually in use that appear in the register

    The diagnostic survey, or a short survey

  3. Share of new uses that went through the approval procedure

    Register

  4. Incidents reported. A rise at first is a good sign: a reporting culture is taking hold.

    Reporting form

  5. Share of uses on unauthorised accounts

    The diagnostic indicator, measured again

  6. How clear the policy is: share of edge cases settled without ambiguity

    The workshop 2 test, run again at 3 months

Which policy format suits us?

As an option, or within twelve-month AI steering: a half-day annual review, because tools change fast. The solidarity rate applies.

Request a quote

What is this policy not?

Who leads this work?

Arnaud Dandoy, PhD in criminology (University of Kent) and law graduate (UCLouvain). More than ten years with public institutions, the United Nations and international NGOs, in Tunisia, Morocco and Haiti. Based in Tunis, working remotely and on site.

Read the full background

Where do we start?

The free hour is enough to see where you stand and what a policy would change. You can also request a quote right away.