Impact Signals #74: A Venezuelan coder's AI missing-persons site fielded 30,000+ reports in 48 hours after twin earthquakes Today: A Venezuelan coder's AI missing-persons site fielded 30,000+ reports in 48 hours after twin earthquakes; Mercy Corps' new agentic AI cuts crisis-analysis turnaround from days to hours in Sudan and Colombia; One typhoon, four warning systems: a Japanese AI stitches the Pacific's fragmented disaster picture together. --- Top Stories **1. A Venezuelan coder's AI missing-persons site fielded 30,000+ reports in 48 hours after twin earthquakes** Jorge Bastidas, a 31-year-old Venezuelan programmer based in Buenos Aires, led a six-person volunteer team that built Desaparecidos Terremoto Venezuela, a no-signup missing-persons platform, in the chaotic days after twin earthquakes struck Venezuela on June 24, 2026 (official and press tallies put the death toll above 3,800, with a submarine cable and much of the country's telecom backbone knocked out). The site runs on Claude Opus 4.8 for matching and coordination, paired with facial-recognition software donated by the Mexican company Lab-Co. It took in more than 30,000 missing-person reports in its first 48 hours, a figure that later climbed toward 45,000 names as the search widened. Rest of World's reporting frames the platform as filling a gap the official response was too slow to close: with cell towers and the country's main undersea cable down, there was no functioning centralized channel for families t…

Impact Signals #73: Mozambique graduates its first AI-assisted drone pilot corps, deploys to five flood zones Today: Mozambique graduates its first AI-assisted drone pilot corps, deploys to five flood zones; KAIST's home-monitoring AI flags cerebrovascular disease risk weeks before diagnosis; Microsoft: AI is cutting post-disaster damage assessment from days to hours. --- Top Stories **1. Mozambique graduates its first AI-assisted drone pilot corps, deploys to five flood zones** Mozambique's National Disaster Management Institute has certified 30 local drone pilots, including 10 newly qualified instructors, completing a 15-month, $967,000 project financed through the African Development Bank's Korea-Africa Economic Cooperation (KOAFEC) Trust Fund. The training programme was delivered by Korean technical partners (Busan Technopark, Hojung Solution and Pusan National University's PNU Drone), who built the underlying AI-enabled disaster-management system: it uses computer vision to help identify people needing rescue and to automatically map damaged infrastructure from aerial imagery, feeding live video to emergency command centres. With training complete, ten of the thirty certified operators are now deploying to five government-identified high-risk zones to run search-and-rescue support, aerial mapping, water-quality monitoring and data collection for early-warning systems. **Why it matters:** this is a template other flood-prone, capacity-constrained countries can copy: rath…

Impact Signals #72: An ESA-backed AI model starts forecasting wildfire spread 12 hours ahead, live inside a commercial platform Today: An ESA-backed AI model starts forecasting wildfire spread 12 hours ahead, live inside a commercial platform; Safe Online commits $8.1M to 30 grantees building AI defenses against AI-generated child abuse material; WMO's new dust storm bulletin puts a number on how much AI has already improved forecasts, and how far it still has to go. --- Top Stories **1. An ESA-backed AI model starts forecasting wildfire spread 12 hours ahead, live inside a commercial platform** FIRE-AID, an ESA-supported project, reached an operational milestone: its first AI-based fire-spread-prediction prototype is now live inside OroraTech's Wildfire Solution platform, the commercial wildfire-management system that already runs daily satellite hotspot detection for fire and civil-protection agencies (including Greece's Hellenic Fire System, which scans the country twice daily). The model combines OroraTech's satellite-based fire archive, terrain, land cover, vegetation and weather data with a sub-daily wildfire-progression dataset, using spatial feature extraction, temporal meteorological encoding, and graph-based modeling to forecast fire spread up to 12 hours ahead from an observed, satellite-detected fire perimeter. ESA is explicit about scope: this is a first operational capability, not a continental-scale simulation tool yet. Users can initialize a prediction direct…

Impact Signals #71: As Typhoon Bavi bore down, China warned amateur AI forecasters they may be breaking the law Today: As Typhoon Bavi bore down, China warned amateur AI forecasters they may be breaking the law; From Mars rover tech to 90-day hunger forecasts: inside the AI systems humanitarians are actually running; South Korea trips its first-ever emergency heatwave warning under a new tiered alert system. --- Top Stories **1. As Typhoon Bavi bore down, China warned amateur AI forecasters they may be breaking the law** As Super Typhoon Bavi approached China's east coast, official channels and state media warned social media bloggers that publishing their own AI-generated storm forecasts may be illegal. The trigger case: a blogger in Shandong province posted that tracking models showed a "90 per cent chance" the storm would push deep into Shandong, a call that diverged from official guidance. Under China's Meteorology Law, public weather alerts run through a centralised release system, and official meteorological stations are the only authorised issuers of forecasts and severe weather warnings. The warning landed in the middle of a real emergency: Bavi made landfall at Taizhou, Zhejiang province, late on July 11 after authorities evacuated more than 1.7 million people in Zhejiang alone and relocated about 34,000 Shanghai residents from high-risk areas (see Pillar 3 for the full disaster picture). Consumer-grade AI weather models are now good enough that individuals can gene…

Impact Signals #70: Lake Victoria AI Saves 450 Fish Cages, WFP Remote-Piloted Aid Trucks, Kerala's Landslide Living Lab Community-scale AI early warning proved itself on three continents this window. An investigation published Friday documents how an oxygen-monitoring system built by the Kenya Marine and Fisheries Research Institute let more than 300 Lake Victoria fish farmers move over 450 cages ahead of a lethal oxygen crash, and Kenya now wants the system at 15 more sites. The World Food Programme unveiled AHEAD, remote-piloted all-terrain aid vehicles built with the German Aerospace Center to take drivers out of the deadliest last mile, while Kerala's Kanichar village switched on India's first disaster living lab with an AI landslide warning system. Today's signal: the early-warning stack that works is cheap sensors, an AI decision layer, and the channel people already use, and the same technology now cuts both ways, as AFP's debunk of an AI-generated Anak Krakatau eruption video shows. --- Top Stories **1. AI Early Warning on Lake Victoria Saved 450 Fish Cages, and Kenya Wants It at 15 More Sites** An investigation published July 10 documents the first proven save by an AI-powered underwater monitoring system on Lake Victoria. Built by the Kenya Marine and Fisheries Research Institute (KMFRI) with Nairobi-based ShoShin Innovation Hub, the system watches dissolved oxygen levels and pushes SMS alerts to fish farmers when readings turn dangerous. When it detected a develop…

Impact Signals #69: UNHCR runs AI across five live humanitarian workflows at once, from displacement forecasting to refugee legal aid Today: UNHCR runs AI across five live humanitarian workflows at once, from displacement forecasting to refugee legal aid; Survey of 2,539 aid workers across 144 countries: 93% now use AI on the job, only 22% work under a formal AI policy; Money Management International opens an AI Lab and a 10-week "AI for Social Impact" fellowship with Anthropic, targeting people who never seek financial help. --- Top Stories **1. UNHCR runs AI across five live humanitarian workflows at once, from displacement forecasting to refugee legal aid** UNHCR published a summit-timed rundown (coinciding with the close of the AI for Good Global Summit in Geneva, July 7-10) of AI tools it is actually running in the field, not piloting on paper. That includes displacement-forecasting models in Somalia, agent-based modeling of return movements in Ukraine, pharmacy-stock prediction in Tanzania, fuel-management analytics in Burundi, and a legal-analysis tool that covers refugee law across more than 200 countries. The agency paired the announcement with an explicit list of the risks it's tracking: algorithmic bias, data gaps in low-connectivity settings, and deepfakes/misinformation increasingly targeting refugee populations. That combination, real multi-country deployment plus a public risk ledger, is the kind of transparency the sector has been asking for. **Why it matters…

Impact Signals #68: Mercy Corps Ships VERA Agentic AI, Google's Flood Forecasts Reach 2 Billion, IIT Mandi's Himalayan Landslide Warnings Cloudera and Mercy Corps shipped VERA, an agentic AI analyst for humanitarian crises that has already cut Sudan food-security analysis from six days to three and saved roughly $2,000 per election-security report in Colombia. The same week, Google published production numbers on its crisis stack, with AI flood forecasting now reaching 2 billion people across 150+ countries, and IIT Mandi switched on a landslide early warning system covering the entire Indian Himalayan Region, delivered over WhatsApp in the middle of monsoon season. Most of it broke around the AI for Good Summit running in Geneva through Friday. Today's signal: humanitarian AI is publishing production metrics, per-report dollar savings, coverage counts, and verified lead times, not pilot promises. --- Top Stories **1. Mercy Corps and Cloudera Ship VERA, an Agentic AI Crisis Analyst With Hard Savings Numbers** Cloudera and Mercy Corps announced VERA (Verified Evidence & Research Assistant) on July 9, an agentic AI solution that automates research, aggregates sources, and produces localized crisis-specific analysis for field teams, running on Cloudera AI Studios and powered by Anthropic's Claude models. The release caps nearly 2.5 years of collaboration and comes with rare production metrics from live operations: up to 90 percent time savings on election security reports in Co…

Impact Signals #67: Google's 2-Billion-Person Flood AI, IRC's Displacement Challenge, Himalayan Landslide Warnings Go Live Google published hard reach numbers for its crisis AI for the first time: flood forecasting now covers 2 billion people in over 150 countries, and its damage models scored 385,000 buildings after Hurricane Melissa hit Jamaica. In India, IIT Mandi switched on a landslide early warning system spanning the entire Indian Himalaya, and the indigenous BharatFS model called a Nashik cloudburst a day ahead, triggering pre-emptive school closures across Maharashtra. Meanwhile the International Rescue Committee challenged the tech industry to scale responsible AI for 118 million displaced people. Today's signal: AI early warning is no longer a pilot category — it is operational infrastructure with reach numbers, live regional coverage, and forecasts that governments now act on before impact. --- Top Stories **1. Google Quantifies Its Crisis AI: Flood Warnings for 2 Billion People, Damage Scores for 385,000 Buildings** Google published a consolidated account on July 7 of how governments and humanitarian agencies use its AI systems for crisis resilience, with unusually concrete reach numbers. Flood Hub now covers more than 150 countries and serves roughly 2 billion people with up to seven-day advance warnings; Public Alerts carries data from authorities in more than 90 countries, and in 2025 Google connected people with crisis information over 10 million times per d…

Impact Signals #66: UN agencies launch flagship playbook for AI in multi-hazard early warning systems Today: UN agencies launch flagship playbook for AI in multi-hazard early warning systems; Google publishes its crisis-resilience scorecard: 2 billion people under AI flood coverage, three new wildfire satellites up today; India's BharatFS AI forecast flagged the Nashik cloudburst a day ahead, and officials acted on it. --- Top Stories **1. UN agencies launch flagship playbook for AI in multi-hazard early warning systems** At the AI for Good Global Summit in Geneva today, ITU, WMO, UNDRR, and IFRC launched "Leveraging AI to Enhance Multi-Hazard Early Warning Systems (MHEWS): A practical resource to support Early Warnings for All (EW4All)." The report, developed by the AI Group of the Early Warnings for All initiative, maps where AI can strengthen each of the four pillars of the early warning value cycle: risk knowledge, detection and forecasting, dissemination, and preparedness to respond. The launch session also showcased the completed AI for EW4All pilot in Malawi, supported by WMO's CREWS mechanism, where MET Norway, ECMWF, and Malawi's Department of Climate Change and Meteorological Services co-developed an AI-based weather prediction system tailored to national needs, with explicit lessons on capacity building and pathways for scaling AI forecasting in Least Developed Countries. Google Research presented its flash-flood prediction work in the same session. **Why it matte…

Impact Signals #65: Britain ships "Extract", an AI that turns paper planning maps into data for every council in England Today: Britain ships "Extract", an AI that turns paper planning maps into data for every council in England; A youth-built typhoon platform, bagyo.app, starts its first city pilot in Naga; Study: federal AI rules leave vulnerable benefit recipients exposed. --- Top Stories **1. Britain ships "Extract", an AI that turns paper planning maps into data for every council in England** The UK Ministry of Housing, Communities and Local Government, working with the Department for Science, Innovation and Technology's Incubator for AI ([i.AI](https://i.AI)), has launched Extract, a tool that reads scanned and handwritten planning documents (PDFs, paper maps) and converts them into structured digital map data. It targets the unglamorous bottleneck that stalls housing delivery: Conservation Areas, Article 4 Directions, and Tree Preservation Orders locked in paper records. Every local planning authority in England gets it free. The numbers are concrete. A task that took a planning officer up to 2 hours per document now runs in about 2 minutes. Within weeks of launch, 50 councils were actively using it and more than 1,000 documents had been processed. Roughly two-thirds of outputs needed only minor edits before use, and one council coordinator reported the extracted data was "100% accurate" on their sample. Development was not a lab exercise: 34 local planning authoritie…