Anthropic Says WHO Team Uses Claude to Cut Ebola Reports to Under an Hour

A daily Ebola situation report that took a full day to prepare now takes under an hour with a Claude tool built by WHO staff, according to Anthropic. The tool is part of a wider effort to use AI in the Bundibugyo Ebola response in eastern Democratic Republic of the Congo, where responders need current information to direct their work.

The partnership was convened by the Coalition for Epidemic Preparedness Innovations, or CEPI. It includes the WHO Regional Office for Africa, known as WHO AFRO, and the DRC’s National Institute of Biomedical Research, or INRB, among others. They are working with Anthropic teams on outbreak data, research evidence and scientific analysis. Those are related jobs, but speeding up a report is not the same as proving that the response has improved.

From field notes to a daily picture

The underlying information starts far from a finished report. Community health workers record illnesses, recoveries and deaths. Health facilities pass case information to districts, often over WhatsApp, while samples go to labs. Districts put their figures into PowerPoint decks, which staff then compile into a situation report. Each handoff takes time before coordinators can see the day’s picture.

WHO AFRO staff built a Claude skill—a set of instructions for a recurring task—to pull case and lab figures from health-zone slide decks. It checks those figures against the previous day’s report, flags changes in trends and summarizes district reports. That targets the compilation work that had kept staff assembling slides late into the night; the reported saving applies to preparing the situation report, not to collecting samples or confirming cases.

Forecasts move closer to logistics

The work goes beyond assembling yesterday’s figures. WHO AFRO staff are also using Claude for analysis and disease modeling. The team can now run multiple disease models and build forecasts to help decide where treatment centers should go. WHO AFRO’s Paul Ouma told Anthropic that time constraints had previously left the team running only one model rather than considering several.

That shift gives logistics staff more possible futures to consider, rather than just a quicker account of the past. Anthropic does not provide a measured comparison of forecast accuracy or show that a Claude-assisted forecast changed a treatment-center decision. The distinction matters in an outbreak: a model can make analysis easier to produce without settling which prediction responders should rely on.

Organizing vaccine work, not choosing the science

There is no approved vaccine for Bundibugyo ebolavirus, Anthropic says. CEPI invited proposals to advance possible vaccines and related research after the strain emerged. It used Claude to build a dashboard for tracking the work teams need to complete. Here the AI’s job is to organize activity across teams, not to produce an approved vaccine.

CEPI is also examining whether samples could help assess potential cross-reactivity between the approved Zaire-strain vaccine Ervebo and Bundibugyo. Polina Brangel, CEPI’s R&D data innovation lead, drew a firm boundary: experts, not Claude, decide which groups of blood samples should be analyzed. She said the tool instead organizes complex proposal data so people can compare it more quickly.

A simpler route through genetic data

At the INRB lab, sequencing produces short fragments of viral genetic code that must be assembled and analyzed. Anthropic says the lab has produced essentially all genomes sequenced from DRC cases in this outbreak. Comparing genomes can help researchers investigate links between cases and changes in the virus that might affect tests, treatments or vaccines.

Claude Science, Anthropic’s workbench for scientists, can take plain-language prompts to assemble a viral genome and build a family tree showing relationships among virus samples. That could lower the need to type specialized commands during analysis. Anthropic describes the capability, but gives no measured reduction in sequencing-to-result time for this outbreak. For now, the clearest quantified gain is in preparing the daily report.