Claude is running WHO's Ebola paperwork. That's not a small thing.

If you haven't been following the Bundibugyo ebolavirus outbreak in the DRC, that's not surprising. It's a rare strain, it's happening in eastern Congo, and it doesn't have the media footprint it deserves given the numbers: 7,672 confirmed cases and 3,699 deaths since the outbreak was declared a public health emergency of international concern on May 17. The case fatality rate is 48.2%. Roughly half of confirmed cases die.

For comparison: the 2014-16 West Africa Ebola outbreak — the worst in modern history — killed about 11,000 people over two years. This one is moving in seven DRC provinces across 63 health zones, and North Kivu has no treatment beds available in its two largest cities.

Anthropic published a feature article in September about how three organizations are using Claude to manage the outbreak response. I want to describe what that actually looks like, because it's a different category of AI use than what most coverage focuses on.

What a situation report actually is

Every outbreak requires a daily situation report — a sitrep. It's a document that summarizes new cases, deaths, recoveries, and trend changes across every affected health zone. For this outbreak: 63 zones across 7 provinces, with data arriving as handwritten notes and PowerPoint decks sent over WhatsApp from community health workers going door to door.

That data has to be collected, cleaned, compared against yesterday's numbers, and compiled into a document outbreak coordinators can act on. Province staff were doing this by hand. It was taking all day.

Tendai Muza, Tamayi Mlanda, and Gianni-Ferrari Donkor Muza at WHO AFRO's emergencies hub in Dakar built a Claude skill — a reusable AI workflow — that extracts case and lab numbers from the zone decks, compares them with the previous day's report, flags trend changes with possible explanations, and summarizes districts' findings. A sitrep that took all day now takes under an hour.

That time is not nothing. Paul Ouma, one of WHO AFRO's data leads, said they could previously run only one disease model per day because of time constraints. Claude lets them run several and build forecasts to help logistics staff decide where to place treatment centers.

What else Claude is doing here

CEPI (the Coalition for Epidemic Preparedness Innovations) is using Claude to organize multi-factor data for vaccine comparison. The critical clarification from CEPI: scientists, not Claude, decide which samples to analyze and which candidates advance. Claude organizes data so experts can compare faster.

There is no confirmed vaccine for the Bundibugyo strain. The only available Ebola vaccine, Ervebo, was developed for the Zaire strain. CEPI is testing whether Ervebo cross-protects against BDBV — a live scientific question. Claude is handling the data organization layer while scientists handle the actual question.

INRB (Institut National de Recherche Biomédicale), the DRC lab that has produced essentially all Bundibugyo genome sequences from this outbreak, uses Claude Science — Anthropic's AI workbench for researchers — to assemble genomes and build phylogenetic trees (family-tree maps showing how the virus is spreading and mutating) using plain-language prompts instead of specialized bioinformatics commands. The trees help trace how the outbreak is moving and identify variants.

Source spread

• Anthropic — The Situation Report [hype]: First-party account from Anthropic of their tool being used in outbreak response. Named personnel and organizations are verifiable. The framing is favorable to Anthropic's product and the article was published by them.

What's real:

• The time-savings are specific and plausible. "All day to under an hour" for a task that is essentially structured data extraction and comparison is exactly the kind of thing LLMs handle reliably. This isn't AI making medical decisions; it's AI doing data work at speed.
• The organizations involved are credible. WHO AFRO, CEPI, and INRB are not startups running demos. These are institutions with accountability for the actual outbreak response.
• Dr. Jean-Jacques Muyembe, INRB's director general and co-discoverer of Ebolavirus in 1976, is quoted: "you beat Ebola by knowing where it is today, not where it was last week." The sitrep pipeline is a direct response to that principle.
• WHO Africa is reportedly using the same Claude pipeline for a chikungunya outbreak. If that's accurate, this is already a template, not a one-off.

What deserves a side-eye:

• We have one source for all of this: Anthropic's own feature article. I'd want to see WHO's own reporting on the deployment or an independent assessment of how the sitrep pipeline is actually working.
• The North Kivu treatment bed shortage (94.7% occupied, none available in Butembo or Katwa) suggests the response is still overwhelmed. AI tools that make administrative work faster don't fix supply constraints or vaccine gaps.
• The BDBV vaccine question is unanswered. CEPI is working on cross-reactivity testing, but that work could fail or take years. Claude organizing data faster doesn't accelerate the underlying science — it just removes a bottleneck in the data layer.

What to know

For everyday readers:

• If you hear "AI in healthcare," this is what the useful version often looks like: not diagnosis, but information management that lets doctors and epidemiologists focus on what requires human judgment.
• The BDBV outbreak will end. But the pipeline — Claude-assisted sitrep generation across health zones — is a template that can run on the next outbreak, wherever it is.
• Anthropic offers Claude Science credits to researchers and Claude for Nonprofits access to frontline health organizations. If you work in a humanitarian context and are wondering whether this is relevant, it is.

For builders:

• The WHO AFRO sitrep pipeline is essentially a structured data extraction + comparison + narrative generation task running on heterogeneous source documents (PowerPoints, WhatsApp messages). It's replicable for any organization running recurring reports from messy, inconsistent source data.
• The design choice that worked: Claude doesn't make outbreak decisions. It does the data layer, flags trend changes, and summarizes. Humans review and decide. That's the right division of labor for high-stakes applications.
• The open question is intake, not analysis. Data still arrives over WhatsApp and in PowerPoints. Claude can process those formats. The manual step is document intake, not the analysis itself.

Further reading

• Anthropic — The Situation Report — the primary account of Claude's use in the DRC outbreak response