How NGOs can use AI tools responsibly
A practical guide for NGO teams on using AI tools: where they help, the main risks, what data to keep out, and a simple policy you can adopt today.
Short answer
NGOs can use AI tools for low-risk tasks such as drafting, summarising public documents, translating and formatting, if a person checks every output. Do not enter personal data, security details or confidential donor information into public AI tools. Write a one-page policy, name who approves new uses, and be open with partners and communities about when AI is used.
Key points
- Use AI for drafts and routine tasks, and always check the result yourself.
- Never paste personal data or security information into a public AI tool.
- Expect errors, bias and invented facts, and plan checks for them.
- Write a one-page AI policy and name an approver.
- Start with one or two low-risk uses and review after a month.
What can AI tools do for an NGO?
AI tools such as chat assistants can produce text, summarise long documents, translate, suggest structures and help with spreadsheets. For a busy team with little time, they can speed up routine work.
OCHA's briefing note on AI and the humanitarian sector says that AI has the potential to improve efficiency in routine tasks like report drafting and data formatting. The same note warns of immediate challenges: misinformation, reinforcement of bias, system opacity, cybersecurity and erosion of privacy. Both points are true at once. Use AI where it saves time and where an error is easy to catch. Do not use it where an error could harm someone.
Which tasks suit AI, and which do not?
Sort tasks by risk before you start.
| Risk | Example tasks | Rule |
|---|---|---|
| Low | Fix grammar, draft a meeting agenda, summarise a public report, suggest a training outline | Use freely; read the result |
| Medium | Draft a proposal section, a donor update or a policy from your own facts | Use with no confidential data; an expert reviews |
| High | Anything with personal data, case files, security or protection information, or decisions about people | Do not use public AI tools |
When you are unsure, treat the task as high risk and ask your manager.
What are the main risks?
- Invented facts. AI can write confident text that is wrong, including fake statistics and sources. Check every number and every reference against the source.
- Bias. AI learns from existing text and can repeat stereotypes about gender, ethnicity or place. Review language that describes people.
- Privacy. Text that you type into a public tool may be stored or used by the provider. Check the provider's terms. Our guide on data protection for NGOs explains what counts as personal data.
- Security. Do not put locations, names of at-risk people or operational details into a public tool.
- Weak local-language output. Quality in Burmese and minority languages varies, so reviewers must be fluent.
- Over-reliance. If staff stop thinking, quality drops. AI should support skills, not replace them.
OCHA also notes that because AI depends on large amounts of data, any AI policy should include relevant data policy. Link your AI rules to your data protection rules.
How do you write a simple AI policy?
A one-page policy is enough to start. Use these six steps.
- State the purpose. AI supports staff; it does not replace human judgement.
- List approved uses. Name low- and medium-risk tasks that are allowed.
- List banned inputs. Personal data, security information, confidential donor and partner documents.
- Require human review. A named person checks facts, tone and bias before anything is sent.
- Name an approver. One manager decides on new tools and new uses.
- Set a review date. Update the policy every six months, because tools change.
Check your donor and partner agreements. Some set their own limits on AI, and laws differ by country. If in doubt, ask a qualified adviser.
How should you be open about AI use?
Trust is the main asset of any NGO, and hidden AI use can damage it. Be open early, and keep the rules simple enough that every staff member can follow them.
Tell partners and donors when AI helped produce a document if their rules require it or if it affects trust. For work with communities, do not use AI to make decisions about individuals, and do not present AI-generated images or voices as real people. Keep a simple log of which tools you use and for what, so you can answer questions later. Include the date, the task, who checked the output and whether any outside data was entered. A log of ten lines is enough, and it helps when a donor or partner asks how a document was produced. It also shows new staff what good practice looks like, and it makes your yearly policy review much easier because you can see which uses were common and which caused problems.
Getting started in one month
- Week 1: Agree the policy and the banned inputs.
- Week 2: Pick one low-risk use, such as summarising public reports, and train the team.
- Week 3: Try a medium-risk use with review, such as drafting a proposal section. Our guide on using AI assistants for proposal drafting shows how.
- Week 4: Meet to review what worked, what failed and what to stop.
Good prompts matter too. Read writing good AI prompts for NGO work for a method.
How EMPO can help
EMPO is a trusted, Yangon-based training and consultancy provider and one of Myanmar's leading names in practical NGO training, with 1,200+ people trained and 15 partner organisations. Our store offers the Applied AI Handbook for NGOs in Myanmar, and we deliver training online and in person, in English and Burmese. See our training services to plan a session for your team.
Sources
- Briefing Note on Artificial Intelligence and the Humanitarian Sector — OCHA, 2024
- Data Responsibility — OCHA Centre for Humanitarian Data
Questions people ask
Can we use AI to write a donor report?
You can use it to draft structure and wording from facts you provide, if no confidential data is shared and a staff member checks every number and claim. Check your donor agreement, because some donors set their own rules on AI.
Is AI safe for translation into Burmese?
It can give a useful first draft, but quality varies and mistakes can change meaning. Have a fluent speaker review anything that will be published or used with communities.
Who is responsible if AI gives a wrong answer?
The person or organisation that uses the output is responsible. AI is a tool, so a named staff member must review and approve what goes out under your name.