Skip to content
Back to blog

Can a nonprofit use Copilot or ChatGPT on its member register? GDPR basis and three safe ways

Håkon Berntsen ·
Can a nonprofit use Copilot or ChatGPT on its member register? GDPR basis and three safe ways

Yes, a nonprofit can use AI on its membership register – but only in a tool the organisation itself has a data processing agreement with, and only for purposes the register already has a legal basis for. Free ChatGPT on a volunteer's personal account may use the content for model training, while Microsoft Copilot Chat signed in with the organisation's work account has “enterprise data protection”. Lists that reveal faith, health, politics or trade union membership should stay out entirely.

This piece builds on our nonprofit AI policy with eight points your board can adopt. Here we go one step further: what applies when the member list itself ends up in the chat window?

Key facts as of November 2026

  • Every item of personal data needs one legal basis per purpose, according to Datatilsynet, Norway's data protection authority (17 March 2026).
  • Membership of a faith community, political party, trade union or patient association can itself be a special category of personal data.
  • ChatGPT Free, Plus and Pro content may be used for training; ChatGPT Business, Enterprise and Edu is not used by default (OpenAI, October 2026).
  • Microsoft Copilot Chat includes enterprise data protection at no extra cost for work accounts, with Microsoft acting as data processor.

Can a Norwegian nonprofit use Copilot or ChatGPT on member data under GDPR?

It depends on the tool, the purpose and the data. Pasting a member list into a chatbot is a new processing operation, and it has to fit a purpose the organisation already has a legal basis for.

After its 2024 project with the Norwegian University of Science and Technology (NTNU), Datatilsynet's sandbox concluded that Norwegian organisations can adopt M365 Copilot – but not everyone, and not for everything. The sandbox report on NTNU and Copilot recommends a controlled roll-out in small steps, starting with selected roles.

What is the legal basis for a membership register?

A legal basis is one of the six grounds in Article 6 of the GDPR that make processing lawful. Datatilsynet's guidance on legal bases (in Norwegian) happens to use a membership association as its worked example:

  • Names: necessary for a contract – the association cannot fulfil the membership agreement without knowing who its members are.
  • Email for marketing: consent – and if a member withdraws it, the association cannot switch to another basis.
  • Postcode: necessary for a contract when used to tell apart members with the same name, and legitimate interests when the purpose is statistics.

The point is that the basis follows the purpose, not the data. In our view the same logic applies to AI. Asking Copilot to draft reminders to members who haven't paid their dues sits naturally under the membership agreement. Profiling members to spot “promising donors” is a new purpose the register is unlikely to cover.

The NTNU report agrees: if AI changes the processing, check whether your legal basis still holds. If it doesn't, Copilot cannot be used for it.

Is membership of an association special category data?

Sometimes, yes. Special category data – often called sensitive data – includes information about religion, political opinions, philosophical beliefs, trade union membership, health and sexual orientation. According to Datatilsynet, processing it is prohibited as a starting point, and you need a specific exemption on top of an ordinary legal basis.

In 2022 the Court of Justice of the EU ruled (case C-184/20) that data capable of revealing sensitive information indirectly also counts as special category data. On our reading, a list of members of a congregation, a party, a union or a patient association reveals faith, politics or health simply by existing.

The nonprofit exemption doesn't cover everything

Article 9(2)(d) allows bodies with a political, religious or trade union aim to process data about their own members, with “appropriate safeguards” and no disclosure to others without consent. Its wording does not mention health. A diabetes or cancer association will therefore typically need explicit consent or another exemption. If in doubt, ask a lawyer first.

How does free ChatGPT differ from a business version on privacy?

Two things: training, and who the contract is with. According to OpenAI's help article on how data is used (as of October 2026), content from ChatGPT Free, Plus and Pro may be used to train its models unless the user turns off “Improve the model for everyone”. A thumbs up or down on an answer can still send the whole conversation for training. ChatGPT Business, Enterprise and Edu are not used for training by default.

The bigger issue is the role. A personal account is a contract between the volunteer and OpenAI. The organisation has no data processing agreement and no control once the volunteer steps down. We explained why the processing agreement matters in our supplier check for nonprofits.

Three safe ways to use AI with member data

1. Ask about numbers, not people

Keep personal data at home. Export totals – members per postcode or age band, say – and let the AI work from those. Note that swapping names for membership numbers is only pseudonymisation. Pseudonymised data is still personal data.

2. Copilot Chat in your own Microsoft 365 tenant

According to Microsoft Learn, Microsoft Copilot Chat (formerly Microsoft 365 Copilot Chat) offers enterprise data protection at no extra cost when you sign in with a work account. A green shield at the top of the window shows the protection is on. No shield means you are outside your organisation's protected environment.

Microsoft's enterprise data protection page states that Microsoft then acts as data processor under its Data Protection Addendum (DPA), and that prompts and responses are not used to train foundation models. Microsoft says its EU Data Boundary covers the EU and EFTA countries, Norway included. Three caveats:

  • Bing web searches fall outside the EU Data Boundary, and for those Microsoft is an independent controller. Keep names out of prompts when web search is on.
  • Anthropic's Claude models are currently excluded from the EU Data Boundary. Choose another model when personal data is involved.
  • Prompts and responses are logged in your tenant. Whatever you paste becomes a new copy of member data that your deletion routine must cover.

If you later move to a full Microsoft Copilot licence, it can surface everything a user has permission to view. Tidy up SharePoint permissions first. If your organisation has no work accounts yet, see our guide to getting started with Microsoft 365 for nonprofits.

3. No AI on sensitive lists

For member lists in faith communities, parties, unions and health associations, the simplest advice is also the best: don't use AI on them. Datatilsynet's sandbox treats a DPIA as the general rule when generative AI processes personal data, and special categories raise the stakes further.

Disclosure: Håkon Berntsen chairs OpenInfo and is CEO of InfoDesk AS, a Microsoft partner that helps organisations with Microsoft 365 and Copilot. The advice above applies whichever vendor you choose.

A checklist your board can adopt

  1. Member data is processed only in tools the organisation holds an account and a data processing agreement for.
  2. Never put member data into personal versions of ChatGPT, Gemini, Claude or similar tools.
  3. Check for the green shield before pasting anything into Copilot Chat.
  4. Web search off, or no names in the prompt, whenever personal data is involved.
  5. AI is used only for purposes the register already has a legal basis for.
  6. No special category data in AI tools without legal advice and a DPIA.
  7. Chat history is deleted after use, or covered by a fixed retention rule.
  8. One board member owns the AI routine, and the decision is minuted.

To work through this with other organisations, the AI day for nonprofits at Microsoft Oslo on Friday 20 November covers safe AI and the questions to ask before adopting new tools. It is free, with 130 places.

Frequently asked questions

Do we need members' consent to use AI on the register?

Not necessarily. What matters is whether the AI use fits a purpose the register already has a legal basis for. Special category data also requires a separate exemption, such as explicit consent.

Is it enough to turn off “Improve the model for everyone” in ChatGPT?

No. It stops training, but gives the organisation no data processing agreement. According to OpenAI, a thumbs up or down can also send the whole conversation for model improvement.

What does the green shield in Copilot mean?

It shows that enterprise data protection is active – you are signed in with a work account and Microsoft's data processing terms apply.

Do we need a data protection impact assessment (DPIA)?

Datatilsynet's sandbox treats it as the general rule when generative AI processes personal data. Genuinely anonymised figures are not personal data.

More to read

Related Articles

Stay in touch

New perspectives, straight to your inbox.

A newsletter about open source, responsible AI and shared knowledge – when we have something to tell.