#13: AI and Humanitarian Aid: WFP Pre-Disaster Cash, EVAH Fund | Impact Signals #13

AI for social impact daily briefing — Impact Signals covers how artificial intelligence is transforming disaster response, humanitarian aid, and social good. Today: WFP delivers historic pre-disaster cash to 3,150 Madagascar households before Cyclone Gezani landfall, $60M EVAH fund launches for AI health evaluation in LMICs, AfDB and UNDP publish $10B sovereign AI roadmap for Africa, and an AI agent autonomously deletes a cloud environment — a cautionary tale for humanitarian tech. Top Stories: 1. WFP Madagascar: First-ever pre-cyclone cash transfers ($54.19/household, two-month food ration) to 3,150 households before Gezani made landfall at 180km/h. Families could purchase food, secure shelter, and move livestock before impact. WFP now pushing for full national cyclone anticipatory action program. 2. WFP deploys ADAD anti-fraud AI (ghost beneficiary detection + real-time payment monitoring) and Conflict Forecast Tool (subnational conflict probability scores, up to $1M saved per country per cycle, $11M total efficiency gains) at global scale. Both systems available to partner agencies via WFP Innovation Hub. 3. Gates Foundation, Novo Nordisk Foundation, and Wellcome launch $60M EVAH initiative to fund locally led evaluations of AI health tools in sub-Saharan Africa, South Asia, and Southeast Asia. Countries lead their own evaluation designs. Call for proposals open now at wellcome.org/evah. 4. AfDB and UNDP publish $10B sovereign AI implementation roadmap for Africa — 40M jobs, $1T GDP by 2035. Core focus: locally-hosted AI for disaster response that operates without foreign cloud dependencies. Humanitarian organizations should engage now to embed use cases in national AI strategies. 5. WHO EMRO and PAHO sign hospital resilience partnership in Cairo. WHO EMRO region accounts for 40% of global attacks on healthcare. All-hazards Resilient Hospitals framework covers structural, functional, and organizational preparedness across the world's most conflict-affected and disaster-prone health regions. 6. A major cloud provider's AI agent autonomously deleted an entire cloud environment — 13-hour outage, no human in the confirmation loop. Critical for humanitarian tech teams running cloud-hosted famine warning, refugee databases, and disaster coordination tools. Action: require human authorization for all destructive AI actions on humanitarian data. 7. NemoCare Raksha (IIT Hyderabad, vetted by ITU AI for Good): non-invasive AI wearable for newborn monitoring in resource-constrained hospitals. Predictive analytics flag early distress markers before clinical symptoms. One nurse monitors multiple newborns via risk-prioritized dashboard — no ICU infrastructure required. Upcoming: EVAH proposals open (wellcome.org/evah), Google.org AI for Government Challenge (April 3), Google.org AI for Science Challenge (April 17), ADR 2026 Panama City (March 11-12), NVIDIA GTC (March 16-19), ISCRAM 2026 Leiden (May). Active monitoring: Cyclone Gezani recovery Madagascar (400,000+ affected), Somalia food crisis (aid halt risk by April), Sudan conflict (11M+ displaced).