Ethics and Confidentiality in the Nonprofit Sector
On this page
Using AI Responsibly
Nonprofits handle some of the most sensitive data in existence: beneficiary contact details, social circumstances, medical information, and data concerning minors. Before integrating ChatGPT into your workflows, you must understand the ethical stakes and establish safeguards tailored to your audience. This lesson gives you the keys to working with confidence, respecting the GDPR (General Data Protection Regulation) and your duty of confidentiality.
What ChatGPT Does With Your Data
On individual plans—Free, Go, Plus, and Pro—your conversations may be used to improve OpenAI’s models unless you disable this option. This also applies to information read by connected apps. For an organization supporting vulnerable people, adjusting this setting takes priority over everything else.
Three levers exist, and they stack. The first is disabling model improvement: in Settings, under Data Controls, turn off the corresponding option, labeled “Improve the model for everyone” in the English interface. Note: even with this off, thumbs-up or thumbs-down feedback on a response can still authorize the use of that entire conversation. The second lever is Temporary Chat, which does not appear in your history and is not used to improve models while active; however, a copy may be retained for up to 30 days for safety reasons. Select non-personalized mode so it uses neither your memory nor your custom instructions: this is the right reflex whenever a topic touches on an individual situation. The third lever is a budgetary choice: on the Business plan, OpenAI does not train its models on your data, and the OpenAI for Nonprofits program makes this accessible to eligible nonprofits at $8 per seat per month with annual billing.
The Golden Rule: Anonymize Systematically
No setting replaces input discipline. Regardless of your configuration, adopt this reflex: never enter personally identifiable information (PII) into ChatGPT. This includes full names of beneficiaries, personal phone numbers and email addresses, home addresses of people you support, social security numbers and medical data, as well as any detailed social situation containing elements that could identify someone.
Rephrasing is simpler than it looks. Instead of writing “Draft a letter for Ms. Dupont, 45, a single mother of three living at 12 Lilac Street in Lyon, receiving RSA benefits,” ask instead: “Draft a template letter for a beneficiary, a single mother, who needs help applying for social housing.” ChatGPT will produce a template that you then personalize with real information outside the tool, in your word processor. The result is identical, but the exposure disappears.
GDPR and Legal Obligations
As a nonprofit, you are responsible for processing the personal data of your members, volunteers, and beneficiaries. Introducing ChatGPT into your practices therefore triggers several concrete obligations, which are better addressed proactively than during an audit.
| GDPR Obligation | Application with ChatGPT |
|---|---|
| Data minimization | Enter only what is strictly necessary; anonymize |
| Informing individuals | Mention AI usage in your privacy policy |
| Record of processing activities | Log ChatGPT usage in your register; on Business, OpenAI appears as a processor |
| Transfers outside the EU | Document the transfer framework: OpenAI’s processing agreement relies on Standard Contractual Clauses |
An Internal AI Charter
We recommend drafting an AI usage charter specific to your organization; two pages suffice, and they prevent many misunderstandings when a new volunteer arrives. It specifies who is authorized to use ChatGPT—staff, volunteers, interns—and what, meaning the types of tasks allowed and those that are prohibited. It then details the how: anonymization rules and mandatory privacy settings for every account. Finally, it establishes two exit principles: all ChatGPT output must be reviewed before distribution, and you decide in advance in which cases you disclose AI usage to your stakeholders.
Bias and Content Verification
ChatGPT reproduces biases present in its training data, and the nonprofit sector is particularly exposed. Watch for stereotypes related to vulnerable populations, as well as paternalistic or stigmatizing phrasing—text referring to “the poor we help” rather than “people we support” sits very poorly with those directly affected. Also beware of invented statistics: the model can “hallucinate” figures with perfect credibility. As for legal or medical recommendations, always verify them with a professional before sharing them with anyone.
Your Turn
Ten minutes to secure your usage. Open your settings, disable model improvement, then review your last three conversations: identify every name, address, or identifiable situation you should have anonymized, and rephrase the prompt as shown in this lesson’s example.
Official sources: How your data is used to improve model performance · Temporary chat in ChatGPT · OpenAI for Nonprofits (OpenAI Help Center) · OpenAI Data Processing Addendum (OpenAI)
Key Takeaways
- On an individual plan, disable model improvement in settings
- Never enter personally identifiable information—anonymize systematically
- The Business plan excludes your data from training; eligible nonprofits access it for $8 per seat per month with annual billing
- Update your GDPR register and privacy policy
- Draft an internal AI charter adapted to your nonprofit
- Always review and verify generated content before distribution