#150: Impact Signals #150 — Epic Pediatric Sepsis Alert Audit, Biohub's $1.8 Billion Open Biology Push, Humanity AI at One Year
AI for Impact Daily Briefing, October 07, 2026
Top Stories
An independent test finds Epic's pediatric sepsis alert caught only 36% of children with sepsis, and 22% of Black children
A retrospective multicenter cohort study published October 5, 2026 in JAMA Pediatrics evaluated the Epic Systems Pediatric Sepsis Early Detection Model across eight emergency departments in one New York health system. It covered 166,943 pediatric ED encounters from January 2024 through June 2025, with 189 confirmed sepsis cases under the Phoenix sepsis criteria. The alert identified 36% of children who met the criteria and fired before sepsis onset in 30% of cases. Positive predictive value was 3%, so most alerts were false alarms, while negative predictive value was above 99%. Sensitivity was 22% for Black children, 20% for children under 5, and 25% in hospitals that were not pediatric-only. The authors concluded the alert "does not support timely or equitable pediatric sepsis screening." Medical Daily reported it contacted Epic for comment and no response appeared in its published material. Why it matters: This is one of the first independent performance checks of a vendor AI alert used on children, and it shows the misses were not spread evenly across patients. The study notes the tool is widely implemented in US emergency departments but had not been independently evaluated. Whether Epic retunes the alert, or health systems change how they rely on it, is not yet known from the sources seen.
US government, Google DeepMind and Meta join Biohub in a $1.8 billion push for open AI biology data
Per Reuters, reported October 7, 2026, the US government, Meta and Google parent Alphabet are joining Biohub, the nonprofit backed by Mark Zuckerberg and Priscilla Chan, to build open datasets for training AI models in biological research. Total investment in the Virtual Biology Initiative reaches $1.8 billion. As reported in summaries of the Reuters story: Meta, Google DeepMind and Isomorphic Labs jointly invest $300 million; the Department of Energy invests more than $500 million over five years in laboratory measurement, modeling and computation; the National Institutes of Health coordinates datasets and repositories built with more than $500 million in earlier federal funding, which Biohub will standardize for AI training; and Biohub committed $500 million in April. The initiative aims to measure how cells respond across far more conditions than scientists have studied and to build predictive models that could shorten drug development timelines that now take years. Why it matters: It moves a private nonprofit effort onto public-sector data and funding, and the datasets are described as open. The disease-impact claims are goals, not results, and nothing in the sources seen shows a deployed health benefit yet. The Axios page was not read in full, so the breakdown rests on Reuters-derived summaries.
Humanity AI marks one year: $186 million in aligned funding across 403 grants, and a new $10 million open call for community-led AI work
Inside Philanthropy reported on October 6, 2026 that Humanity AI, the $500 million grantmaking collaborative led by the MacArthur Foundation and Omidyar Network with funders including Ford and Mellon, says it and its funding partners committed more than $186 million in aligned funding across 403 grants to 341 grantees in its first year. It has also awarded 10 grants totaling $8 million in pooled funding to 12 grantees. In September 2026 it opened its first open call, setting aside $10 million for community-led efforts that shape how AI is built, governed and used. Beth Goldberg, formerly of Google's Jigsaw, became executive director in July 2026. The piece also describes a Kentucky digital town hall pilot that drew more than 8,000 residents. Why it matters: It is the first full-year ledger of one of the largest philanthropic efforts on AI and public agency. The article is plain about the limits: $500 million is small next to tech-industry AI infrastructure spending, which reportedly exceeds that each month, and federal regulation looks unlikely under current policy. Whether community-led projects change how AI systems are actually built is not yet shown by anything in the source.
EmberEye raises £1.5 million for tree-canopy infrared sensors that flag wildfires within a minute of ignition
Cambridge-based EmberEye announced £1.5 million on October 7, 2026 to expand deployment in the UK and US. Its battery-powered ground-level infrared sensors detect the thermal signature of flames, not video or audio, and an AI platform assesses wildfire risk and flags high-risk areas to land managers and responders through a cloud dashboard. The founders are electronics engineer Enrico Miraglia Del Giudice and Stanford PhD student Kyle Nisbet. All performance figures are company claims: a 100% detection rate across 20 field demonstrations (with the National Trust, Surrey Fire and Rescue Service, the City of London Corporation and the Ministry of Defence), and a California trial with Marin County Fire Department that detected a fire in under 60 seconds from more than 100 metres away. Why it matters: Detection speed is the claimed advantage, but the evidence so far is 20 supervised demonstrations reported by the company, with no independent evaluation or live-season false-alarm rate seen. The funding pays for wider deployment, which is where those claims will be tested. Investor names were not seen in the sources.
Wales funds "Sgil AI": £499,998 to train 625 charities and pay 12 young people as trainee service designers
ProMo Cymru secured £499,998 from The National Lottery Community Fund for Sgil AI, a five-year program announced October 7, 2026. It will train 625 third sector organizations across Wales using a tiered AI Maturity Model (Aware, Experimenting, Confident, Creative), create 12 paid Trainee Service Designer roles for young people, and fund a cohort of about 10 organizations to design and test AI uses with the communities they serve. Confirmed co-design partners are the Welsh Refugee Council, Mind in Gwent, Stori, and CVSC (Community and Voluntary Support Conwy). The program cites the 2025 Charity Digital Skills Report: 76% of charities were using AI and 60% were concerned about its effect on service quality and data privacy. Why it matters: The funding is built around the gap between high charity AI adoption and low confidence in using it safely. Tools and findings will be published openly in Welsh and English, with independent evaluation at the midpoint and the end, so the results will be checkable.
Upcoming Events & Opportunities
Active Disaster Monitoring (GDACS/OCHA)
- Eastern Pacific:** Tropical cyclone TWENTY-E-26, GDACS Orange notification dated October 7, 2026. Population affected counts were not read from the feed; response status not verified. Source: gdacs.org.
Sources: See individual stories above for full attribution.