#113: Impact Signals #113 — A school-by-school hazard model went national today, and the point is to stop cancelling classes across whole provinces
AI for Impact Daily Briefing, August 26, 2026
Top Stories
A school-by-school hazard model went national today, and the point is to stop cancelling classes across whole provinces
Source URL: gmanetwork.com The Philippine Department of Education launched Project LIGTAS+ nationally in Makati City today, 26 August, with Education Secretary Sonny Angara presiding. LIGTAS+ stands for Learning Institution Geohazard Tracking and Assessment for Safety. It was built by DepEd's Disaster Risk Reduction and Management Service with the Education Center for AI Research, and it had been running as a pilot since 8 May under the M7X Ready School Program, so today is the national rollout of a tested system rather than a first reveal. The platform combines satellite imagery, geospatial analytics, interactive multi-hazard mapping and historical hazard data to generate a risk profile for an individual school against typhoons, floods, earthquakes, volcanic activity and landslides. Flood intelligence is drawn from satellite SAR data, and the AI weather component produces an outlook up to 10 days ahead. The operational goal Undersecretary Malcolm Garma described is granularity: more science-based and data-based decisions on suspending and resuming classes at the level of the individual school, instead of the blanket area-wide suspensions that keep learners home when their own campus was never at risk. DepEd published no school count, learner count or budget figure at launch. Why it matters: Blanket class suspension is a resilience decision that costs learning days at scale, and until now the unit of that decision in the Philippines has been the city or the province. LIGTAS+ moves the unit to the campus, which changes who has to justify a closure and on what evidence. The 10-day forecast horizon is the operational detail to watch, because a system that can only warn 12 hours ahead cannot pre-position anything. What is verifiably next is the first typhoon season run under national coverage, and whether DepEd publishes the school-level coverage numbers it did not release today.
The AI was right 99 times out of 100. Doctors stopped using it anyway, and the ER got no faster
Source URL: nature.com Nature Medicine published on 19 August a prospective, real-patient evaluation of SHAKED, a clinical decision support system built on multiple large language models, running inside a tertiary emergency department. The design is a DECIDE-AI stage 1 study, the early-clinical-evaluation standard, and it covered 1,138 patients across four weeks in two parallel units, one working with SHAKED and one on routine rotation. The authors are Liron Leibovitch, Adi Ahituv, Alon Gorenshtein, Dvir Aran, Moran Sorka, Keren Miron and Shahar Shelly, working from Rambam Health Care Campus and the Technion in Haifa, with co-authors at Beth Israel Deaconess Medical Center in Boston and the Mayo Clinic in Rochester, Minnesota. The safety and quality results were strong. No adverse events were detected. Expert reviewers rated 99 of 100 sampled outputs clinically appropriate. Physicians particularly favoured the system for radiology consultations. The result that matters is the one nobody designs for: clinical adoption fell from 68 percent to 30 percent over the four weeks, which the authors attribute to workload-sensitive disengagement, meaning clinicians dropped the tool exactly when the department got busy. Emergency department length of stay was 4.9 hours in both wings, unchanged. The authors state plainly that these findings do not justify clinical deployment of AI decision support at this stage, and that sustained clinician engagement, rather than algorithmic accuracy, may be the key barrier. Why it matters: This is one of the few prospective, in-workflow measurements of a hospital LLM assistant rather than a benchmark score, and it separates two things that are usually reported as one. Accuracy was not the failure point: 99 of 100 outputs passed expert review and no patient was harmed. Usage was. A tool that is used by 30 percent of shifts at week four cannot move a department-level metric, which is why length of stay did not move. The finding is newly measurable evidence that the decisive variable in clinical AI is whether busy staff keep reaching for it, and the authors have put a number on how fast that can decay.
Kenya just put a price tag on maternal-health AI: tenders opened 26 August for the modules that would follow a pregnancy end to end
Source URL: techweez.com Kenya's Digital Health Agency has floated procurement for the reproductive, maternal, newborn, child and adolescent health modules of Taifa Care, the national digital health platform. Techweez reported the plan on 25 August. The tender itself, DHA/ONT/02/2026-2027, covers design, development, integration, maintenance and support of modules for antenatal care, delivery and postnatal care, nursing, immunisation, adolescent health, basic and comprehensive emergency obstetric and newborn care, and maternal and perinatal death surveillance and response. Tender documents were listed as available from 26 August 2026 at 12:00, with a closing date of 9 September 2026 at 10:00. Taifa Care is six interconnected systems under a single brand. The AI functions described are specific and unglamorous: flag a missed antenatal visit, forecast blood demand so a bank is stocked before a haemorrhage, identify duplicate records, and help nurses triage which cases need attention first. The platform is also meant to connect births and deaths to both the health authorities and the civil registry, which is where a maternal-death surveillance system either works or quietly does not. Techweez's own assessment is the caveat worth carrying: data only matters when it leads to a properly staffed and funded response, and the tender documents carry no published budget figure, rollout schedule or facility count. Why it matters: The gap between a government announcing an AI health strategy and a government paying a vendor to build specific modules is where most national digital health plans stall, and Kenya has now crossed it with a named tender, a stated scope and a dated deadline. What is newly true is that the maternal-death surveillance component is inside the procurement rather than deferred, so the same system that forecasts blood demand is the one that records whether a woman died. For anyone building in this space, the 9 September closing date is the concrete, verifiable next step for this story.
The EIB writes its first climate-insurance cheque, and 210,000 Ethiopian farmers are the test
Source URL: eib.org EIB Global, the development arm of the European Investment Bank, is putting a grant of up to 4 million euros, about 4.7 million dollars, into the World Food Programme to develop and roll out climate-risk insurance for Ethiopian smallholders. Around 210,000 farmers are expected to be covered by microinsurance products against crop and livestock losses tied to adverse weather and natural disasters. The EIB says this is the first climate-risk insurance project it has backed. The announcement was made at the ongoing UN Convention to Combat Desertification COP 17 in Mongolia, and the item entered our candidate pool on 26 August at 09:50 UTC. The mechanism is index-based, also called parametric: payouts trigger off a measured index rather than a farm-by-farm loss assessment, which is what makes covering 210,000 scattered smallholdings arithmetically possible in the first place. WFP will design and pilot the scheme with private insurers and roll it out through selected rural finance institutions under the third phase of Ethiopia's Rural Financial Intermediation Programme, with technical assistance, capacity building and awareness raising attached. WFP will initially administer the Premium Guarantee Fund before handing management to the Development Bank of Ethiopia. The grant complements an existing 110 million euro EIB credit line to RUFIP III. Why it matters: Ethiopia is one of the countries where WFP has run satellite-index insurance longest, so the news is not the concept but the balance-sheet shift: a European public bank is now underwriting climate-risk transfer directly rather than only lending to the rural finance system around it. What is verifiably next is the handover of the Premium Guarantee Fund from WFP to the Development Bank of Ethiopia, which is the point at which the scheme either becomes a domestic instrument or stays a donor programme with a local address.
Upcoming Events & Opportunities
Adaptation Fund Climate Innovation Accelerator (AFCIA), delivered by UNEP CTCN (Funding)
- Amount: up to USD 150,000 in technical assistance support per project (technical assistance, not a cash transfer); up to 10 projects selected
- Deadline: 7 October 2026
- Eligibility: entities based in developing countries with an officially nominated National Designated Entity to the UNFCCC Technology Mechanism, covering government institutions, universities, research centres, NGOs and private-sector entities. Least Developed Countries and Small Island Developing States especially encouraged. This call is the Asia Pacific regional window.
- Apply: ctc-n.org
- Apply: ctc-n.org
- Kenya Digital Health Agency, Tender DHA/ONT/02/2026-2027 — design, development, integration, maintenance and support of RMNCAH modules in Taifa Care; documents listed as available from 26 August 2026 at 12:00, closing 9 September 2026 at 10:00. Deadline seen on a tender aggregator, not on the agency site; confirm at dha.go.ke or tenders.go.ke before relying on it. tenderyetu.com
- Adaptation Fund Climate Innovation Accelerator, Latin America and the Caribbean window — same USD 150,000 technical assistance structure; this window closed 18 August 2026. Closed, listed for completeness. ctc-n.org
- Moderna Foundation Grant Program (United States) — surfaced in the 2026-08-25 candidate pool via fundsforNGOs; amount, eligibility and deadline not verified against the funder page. Do not treat as open until confirmed.
- African Engineering Innovation, GBP 85,000 — surfaced via ICTworks on 2026-08-24; deadline not verified against the funder page.
- UNICEF Venture Fund — rolling themed open calls offering equity-free funding to early-stage startups building open-source AI and machine learning for children; each themed call opens and closes independently, so there is no single deadline to quote. Verify the currently open theme before applying.
Active Disaster Monitoring (GDACS/OCHA)
- Bhotekoshi river flash flood, Rasuwa district, Nepal (Nepal-China border):** at least 8 bodies recovered per Al Jazeera, with local police in Bagmati province putting the toll at at least 17; more than 100 Indian nationals reported missing; 26 Nepal Police and 7 Armed Police Force personnel missing. Settlements and infrastructure damaged at Syapru Besi and Timure in a district of roughly 50,000 people. Across the border the surge hit China's Gyirong port in Shigatse, cutting roads, communications and power. active search and rescue; Nepali officials are examining whether a glacial lake outburst flood or a landslide-dam breach in Tibet caused the surge. Cause not yet established.
- M6.0 earthquake, 215 km NNE of Lospalos, Timor-Leste:** offshore, no significant population impact reported in the feed monitoring
- M6.0 earthquake, 33 km SSW of Honcho, Japan:** populated region, no major damage reported in the feed monitoring
- M5.8 earthquake, 4 km N of Toride, Japan (Greater Tokyo area):** shallow event under a dense population centre, included because it clears the M5.5-in-a-populated-area bar; no major damage reported in the feed monitoring
- M6.7 earthquake, 31 km NW of Aniso, Peru:** strong event, impact assessment not in the feed monitoring
- Tropical Cyclone SAUDEL-26 (ongoing):** 7.679 million people exposed to Category 1 winds (120 km/h) or higher ongoing, GDACS Orange
- Lao PDR floods (ongoing since July 2026):** per-country situation page; no consolidated casualty figure available in the feed ongoing
- Note: only major or ongoing-major disasters are featured; low-severity GDACS Green alerts are excluded per the major-only bar.
Sources: See individual stories above for full attribution.