- Use AI to speed up monitoring and analysis, with systematic source verification.
- Recognise the new risks created by AI: disinformation, deepfakes, targeted phishing.
- Protect security information and people, in line with the duty of care.
This module complements the 12 weeks of the programme. It adds no exam: certificates and the diploma are earned as before. Tools are described as known in November 2026; they change quickly.
1Open-source monitoring with AI
A security manager reads articles, press releases, social media posts and reports from other organisations every day. AI can:
- Sort a large volume of articles by criteria (area, incident type, date).
- Translate sources in other languages quickly to grasp the essentials.
- Summarise the facts, distinguishing what is confirmed from what is reported.
- Extract date, location and incident type into a table for analysis.
But a summary can distort: confuse two places with the same name, turn a rumour into a fact, get a date wrong. Every incident kept in a report is checked in the original source and, where possible, cross-checked with a second independent source or the local network.
2Incident and trend analysis
From a well-maintained incident database, tools (spreadsheet, statistical software, mapping) show trends: monthly change, high-risk areas and times, incident types. Generative AI helps write the formulas, produce charts and draft a first analysis.
Important limits:
- An incident database is incomplete: many incidents are not reported, especially those affecting national staff or poorly covered areas.
- A model does not "predict" an attack: it shows probabilities based on the past. Context analysis (actors, intentions, capabilities) remains human.
- Risk levels and measures (suspension of movements, relocation) are decided by the designated managers, according to the security plan.
3Disinformation, fake content and phishing
AI has also created new risks for humanitarian organisations:
- Disinformation: fake press releases, false accusations against an NGO, amplified rumours that can trigger threats against staff.
- Generated images, videos and voices (deepfakes): a fake video of a manager or a fake voice message from a director asking for an urgent transfer.
- Targeted phishing: well-written emails, in the local language, impersonating a donor or a colleague.
Measures: a verification procedure through a second channel for any payment request or change of bank details; reputation monitoring; a rumour response plan with the communication team and community focal points; regular staff training.
4What must never go into an AI tool
- The routes, schedules and positions of teams, convoys and guesthouses.
- The names and contact details of staff, community contacts, informants and partners.
- Detailed incident reports, kidnapping, crisis management and hostage files.
- Security and evacuation plans, radio call signs, passwords.
Even a professional account does not justify sharing this information without a real need and the manager's agreement. When in doubt, do not share. A monitoring summary built from public sources can, however, be prepared with AI, provided no internal information is added to it.
5Duty of care and accountability
The duty of care requires the organisation to take reasonable measures to protect its staff, including national staff and partners. AI is an analysis aid; it does not reduce this responsibility.
- Document the sources used and the share of the analysis done with AI.
- Avoid bias: monitoring that reads only English or French sources can miss local signals.
- Share useful information with other organisations through the agreed coordination channels, without exposing people.
- Do not transfer to local partners risks they cannot manage.
Exercise: evaluating AI-assisted monitoring
Situation. In one week, the tool sorted 200 articles and kept 35 as relevant incidents for your area. Your analyst checks all 35: 28 are confirmed, 7 are outside the area or duplicates. The analyst also rereads a sample of 40 discarded articles and finds 4 relevant incidents.
Questions. 1) What share of the incidents kept is confirmed? 2) What share of the discarded articles reread contained a missed incident? 3) Conclusion?
Answer key. 1) 28 / 35 = 80%. 2) 4 / 40 = 10%: across the 165 discarded articles, around fifteen missed incidents can be expected. 3) The tool saves time but misses incidents: keep sample re-reading, revise the sorting criteria (place names, languages), and never rely on automatic sorting alone for a security decision.
Key points
- AI sorts, translates and summarises monitoring; every incident kept is checked in the source and cross-checked.
- Trends support analysis, but no tool predicts an attack; the decision stays with the designated managers.
- Disinformation, deepfakes and phishing: second-channel verification and staff training.
- Never routes, positions, names or security plans in an AI tool.
- The duty of care remains in full: document, avoid bias, protect partners too.
Self-assessment quiz
6 self-assessment questions, corrected immediately. They do not count towards the exams.
The Mini MBA AI and Humanitarian Action (Claude and Gemini) covers these topics in depth over 12 weeks.