SAFE AI · EU AI Act · Humanitarian AI
Almost everyone in our teams already uses AI. Very few humanitarian organisations have written down what is allowed, what is not, and which data must never leave the organisation. Here I show what the EU AI Act and the SAFE AI framework actually require, then apply them, requirement by requirement, to two real MEAL tools.
The recording of the 26 September 2026 session, in French. You can pause at each demonstration and go straight to the part that concerns you.
At the end of the session, every participant could rate what they had just attended. MEAL officers, MEAL coordinators, protection officers, consultants, statisticians: here is what they took away.
38 participants responded, on a scale of 1 to 5. Percentages count scores of 4 and 5. Anyone who responded more than once is counted once, with the average of their responses.
Opti is a tool that is revolutionising MEAL. Thank you, Thomas.
A great presentation. We would very much like a future session on choosing SMART indicators.
I learned a lot. Thank you.
Regulation (EU) 2024/1689 regulates uses of AI rather than a technology: the more a use can harm people, the heavier the obligations. It makes no humanitarian exception. An NGO is almost always a deployer: it uses tools designed by others, and its obligations cannot be handed over to the provider.
Some practices have been banned since February 2025, including exploiting vulnerabilities and social scoring. The transparency obligation has applied since 2 August 2026. High-risk rules, which directly cover access to essential aid, emergencies and migration, were postponed to 2 December 2027 by the July 2026 AI omnibus. The same omnibus reworded Article 4: organisations must now support the development of AI literacy among their staff, and be able to show what they have done.
A national organisation with no link to the EU is not directly subject to it. Its European donors and partners are, and they pass these requirements on through their contracts, as they did with the GDPR. The details, article by article: the EU AI Act explained for humanitarian organisations.
SAFE AI (Standards and Assurance Framework for Ethical Artificial Intelligence) is the AI governance framework designed for humanitarian action, published as version 1.1 in May 2026 by the CDAC Network, the Alan Turing Institute and Humanitarian AI Advisory. It is voluntary, and goes further than the law on humanitarian principles and the role of communities.
It classifies each use case, not each tool, into three risk levels, and takes every project through four stages, each closed by a decision: continue, redesign, pause or stop. It recognises responsible refusal as a valid outcome: not using AI is a SAFE AI result in its own right. Its central document is the transparency card, published before going live. The full framework: SAFE AI, the humanitarian AI framework.
Principles alone are not enough to judge a tool. During the session, I apply both texts to two real uses, on Opti', the MEAL platform I designed. I say so at the start: it serves as evidence, and each demonstration also shows its limits.
Each method page comes with a pre-filled SAFE AI transparency card, editable in Word, so you can see what the document looks like once completed.
The guide covers the essentials of the session. The other resources go further on each part, at your own pace.
The key points in five ideas, the 2026 landscape, the AI Act, SAFE AI, the two case studies, a grid to classify a use in five questions, and three steps for Monday.
Download the guide → Free trainingNine steps, about an hour, written from the framework's own text, with an exercise to classify your own uses. Included with the free Opti' account.
Start the training → GuideWho is concerned, what is banned, high risk, transparency, AI literacy and the timeline.
Read the guide → GuideThe right to know, humanitarian principles, the three risk levels, the four stages and the framework's tools.
Read the guide → Demonstration 1How an XLSForm is built and checked before fieldwork, with its SAFE AI transparency card.
See the method → Demonstration 2How a narrative report is written and checked figure by figure, with its SAFE AI transparency card.
See the method →The same session has been given twice, and each edition built on the one before.
The place of AI in analysing monitoring and evaluation data: what it does well, structuring reasoning and writing, and what it should not do, the calculations.
Rebuilt around the AI Act and SAFE AI, with two demonstrations classified under both texts. This is the recording on this page.
Opti' covers the whole MEAL chain: XLSForm questionnaires, data collection, analysis and narrative reporting. Calculations are done by fixed rules, and what the AI writes is checked automatically before it reaches you. The account is free, and the SAFE AI training is included without using any tokens.
I designed it after eight years in the field. If your organisation needs a tool built on the same principles for its own uses, that is what my custom AI tools.
I can deliver it for your teams, your leadership or your network, adapted to your tools and your cases. September's participants also asked for follow-ups: data protection in crisis settings, choosing SMART indicators, using AI in MEAL data management. These are topics I can build with you.
For your MEAL or programme teams: what AI can do for them, within what framework, and how to classify your own uses under the AI Act and SAFE AI.
For a leadership team opening the AI question: the obligations, the sector references, and the first decisions to make. It is often the starting point for AI advisory.
For a network, a cluster, a conference or an internal event: a talk on ethical AI in humanitarian work, with a live demonstration, adapted to your audience.
No. SAFE AI is a voluntary standard for the humanitarian sector. It does, however, give donors a basis for asking for evidence proportionate to the risk of a use, and that is often how it enters organisations.
Directly, if the organisation is established in the European Union, or if the output produced by the AI is used there. A national organisation with no link to the EU is not subject to it, but its European partners pass its requirements on through their contracts. The details: the EU AI Act and humanitarian organisations.
The AI Act is a European law, based on four risk levels. SAFE AI is a voluntary framework specific to humanitarian work, based on three risk levels, humanitarian principles and community participation. The session shows how to classify the same use under both.
Yes. The replay, the guide and the two reference pages are freely available. The SAFE AI training only requires a free Opti' account.
Tell me who your audience is, and what you want them to take away. I will suggest a format.
