#158: Impact Signals #158 — Zimbabwe's DawaMom AI Midwife, UNAM's Regresa for Mexico's Missing Children, Georgetown's 53 Percent Budget Error, Disability AI Studies Without Disabled People
AI for Impact Daily Briefing, October 17, 2026
Top Stories
DawaMom, a WhatsApp AI Midwife Grounded in Health Ministry and WHO Guidance, Reaches 20,000 Women in Zimbabwe
RFI reported on October 17, 2026 that DawaMom, an AI-powered WhatsApp chatbot for pregnancy information built by Zambia-based startup DawaHealth, has supported more than 20,000 users in Zimbabwe and aims for 50,000 by the end of 2026. It uses retrieval-augmented generation and answers only from a verified database of Health Ministry guidelines, WHO recommendations, medical school curricula and hospital protocols. More than 25 health workers translated and validated the content into Shona, Ndebele and Tonga, the bot can reply by voice in a Shona accent, and it automatically flags high-risk women for follow-up with licensed health workers. Basic access is free; paid tiers add personalized monitoring and home visits. Why it matters: Zimbabwe had just over 22 health workers per 10,000 people in 2022, under half the WHO benchmark of 45, and this service puts a closed, ministry-sourced knowledge base in front of tens of thousands of women in three local languages. The next checkable number is the 50,000-user target by year end; so far the evidence is user testimony, with no independent outcome evaluation reported.
UNAM's Regresa Ages the Faces of Mexico's Missing Children, Built on Mexican Faces Because Imported Software Misreads Them
Agence France-Presse profiled Regresa ("Return") on October 16, 2026, a project at UNAM led by anthropologist Ana Itzel Juárez that uses anthropometry to model craniofacial growth and generate updated images of children on old missing-person posters. The tool runs on its own image bank, built from hundreds of photos and data from 75 Mexican volunteers, because open-source face databases poorly reflect typical Mexican features; AFP reports most facial software used by prosecutors' offices is not suited to Mexican facial traits. More than 134,000 people are being sought in Mexico. CNN first reported the project in March 2026, so this is a new profile, not a launch. Why it matters: The finding is about representation: the facial software already in some prosecutors' offices was not built on faces like the ones it is asked to match, so UNAM built its own reference set. No accuracy figures and no formal hand-off to authorities have been reported, so the tool's value to the open cases is not yet demonstrated.
Georgetown Report: ChatGPT Put a California County's Public Safety Spending 53 Percent Above the County's Own Budget
Georgetown University's Beeck Center for Social Impact and Innovation released "Making Government a Trusted Source in AI Systems" on October 14, 2026, StateScoop reported. ChatGPT gave a public safety spending figure for Mendocino County, California, 53 percent higher than the county's own budget tool; the report suggests the county's JavaScript-based dashboard made the real number hard for AI systems to retrieve. A plain Larimer County, Colorado page on special-event permit timelines was relayed accurately, and the report found that security and bot-management tools can unintentionally block legitimate AI crawlers from public information. Why it matters: The error traced back to a public budget published in a format the assistant could not read, which moves part of the accuracy problem onto something governments control: how their own pages are built. The report recommends tiered crawler management that admits verified AI crawlers, named ownership of the fixes, and an annual review.
A Ten-Year Review of 93 AI Health Studies for People With Disabilities Finds Only 6 Involved Disabled People
A scoping review published October 15, 2026 in PLOS Digital Health by Université de Montréal researchers, led by Soutongnoma Safiata Kaboré, screened 4,233 records and kept 93 studies (2015 to 2025) of AI health applications for people with motor, sensory, cognitive, communication-related and psychosocial disabilities. About 60 percent referenced equity, but only 12 (12.9 percent) implemented concrete equity measures and only 6 (6.5 percent) involved people with disabilities through co-design or user validation. 73.1 percent of studies came from high-income economies; Africa accounted for 1.1 percent. Why it matters: The roughly 1.3 billion people with disabilities worldwide are the intended users of these tools, yet a decade of research almost never put them in the room, and the evidence base comes overwhelmingly from high-income settings. The authors call for diverse samples, funding for co-design, and equity evaluation built in from the start of a project.
Daily AI Use Among US First Responders Nearly Tripled in a Year, to 32 Percent, in Verizon's 1,035-Person Survey
The 2026 Verizon Frontline Public Safety Communications Survey, released October 13, 2026 and conducted by Lexipol Media Group across 1,035 US public safety professionals, found daily AI use reached nearly 32 percent, about triple the 2025 level. Fifty-six percent of agencies expect daily AI use within three years, daily use of drones and robots rose to 26 percent from 15 percent, and more than 77 percent named network resilience and reliability as their main operational requirement. This is a vendor-sponsored, self-reported survey, and its figures have not been independently verified. Why it matters: Routine daily use, not pilot interest, roughly tripled in one year, while the field's top requirement remains a network that stays up, during the disasters when networks fail most. No public evaluation of how that AI use performs in emergency work has been reported alongside the adoption figures.
Upcoming Events & Opportunities
AccessAbilities Expo 2026 (8th edition)
- Opens October 19, 2026 (date per The National Law Review; full dates not confirmed)
- Location: Dubai, UAE
- A regional expo on accessibility and assistive technology for people with disabilities.
- Details: news.google.com
World Landslide Forum 2026
- November 23 to 27, 2026 (dates per ET Now)
- Location: India
- The global forum on landslide risk reduction, including monitoring and early warning.
- Details: news.google.com
Active Disaster Monitoring (GDACS/OCHA)
- Guam and Northern Mariana Islands:** Tropical Cyclone CHAMPI-26, GDACS Red alert, maximum wind speed 259 km/h; about 175,000 people in Category 1 or higher winds. Ongoing. Source: gdacs.org.
- Balochistan, Pakistan:** M5.7 earthquake 48 km WNW of Khuzdar on October 17 (05:06 UTC), USGS PAGER Yellow; damage and casualty figures not yet confirmed. Source: usgs.gov.
- Mexico (Eastern Pacific):** Tropical Cyclone SIMON-26, GDACS Orange, maximum wind speed 250 km/h during its lifetime; winding down as of October 16. Source: gdacs.org.
- Note: only major or ongoing-major disasters are featured; low-severity alerts are excluded.
Sources: See individual stories above for full attribution.