Miliband: AI Can Make Aid Proactive as Funding Falls by Half

Humanitarian groups are betting that AI can help them act before disasters strike, just as the money to respond after them dries up. At a Sept. 28, 2026, side event during the UN General Assembly, David Miliband of the IRC described using Claude to forecast pneumonia hotspots and find unvaccinated children in Burkina Faso. CEPI, the World Food Programme and the governments of Kenya and Rwanda also detailed deployments, alongside new access commitments from host Anthropic.

---

• 1 big thing: IRC uses Claude to anticipate pneumonia and locate zero-dose children after absorbing $500M in US cuts
• Ebola situation reports cut from days to 12 minutes, CEPI's deputy CEO said, as the DRC outbreak tops 3,600 deaths
• WFP and Anthropic build "Plenty," an open-source hunger early-warning system, as three famines loom
• Kenya signs a sector agreement with Anthropic; Rwanda cites 30-plus AI use cases
• Anthropic offers a 90% discount for nonprofits in low- and middle-income countries and highlights its $200M research fund

---

Miliband's Pitch: Anticipate the Crisis

Why it matters: Miliband said humanitarian aid has fallen 50% in the last year and a half while 300 million people depend on it to survive. His answer is a sector that targets need earlier and more cheaply.

The basics: The segment opened with an IRC video from Nigeria. It featured Mariam, a nutrition officer who sees up to 100 malnourished children a day and records their data by hand on cards.

"All the paperwork and documentation that need to be done slow down the aid that the child is receiving," said Mariam, nutrition officer, IRC.

"Most of the data that are gathered, maybe about 50%, is not ultimately used because of how long it takes," said a narrator in the video.

Where things stand: Tino Cuéllar, Chief Global Affairs Officer, Anthropic, opened the fireside chat by asking Miliband about the UN's relevance. Miliband invoked Harry Truman's 1945 "right makes might" charge. He argued that international power has become diffuse while national power is concentrated.

"Only 7% of the world's population now live in a liberal democracy," Miliband said. He tied that shift to 65 wars and 122 million people displaced by conflict and disaster.

The money picture is stark. Miliband said the IRC grew from $500 million to $1.5 billion but absorbed $500 million in US government cuts over 18 months. He said "impact, innovation, cost-effectiveness" guided its response.

"It needs to become proactive where it is currently reactive," he said of the humanitarian sector.

The AI use case: In Burkina Faso, Miliband said, Claude synthesizes information to flag whole communities likely to see high pneumonia prevalence, not just individual patients.

"We're using disparate datasets to identify the communities that have zero-dose kids," Miliband said. That data then shapes mobile vaccination routes. The longer-term aspiration is to forecast local conflict from displacement data.

The other side: Miliband drew clear limits on the technology.

"Negotiating with terror groups to get access for humanitarian aid workers — Claude can't do that, but people can do that," he said.

"What I find promising about this example we're talking about is that it's a good, compelling example of human-machine collaboration," said Cuéllar.

Decisions and stakeholder impacts: Miliband called on donors to fund integrated rather than siloed grants. He said the IRC will share its findings with other organizations. He cited Dioptra, a cost-effectiveness collective in which 9 NGOs share input data, and urged UN agencies to contribute their own data.

He also criticized the end of USAID education funding in Afghanistan. "It's pure performative cruelty that's leading to 300,000 kids in Afghanistan who were dependent on a USAID grant for education, boys and girls, to lose that education," he said.

What's next: Miliband cited a cost of $2 per vaccination. He called 45 million acutely malnourished children and 1.5 million childbirth deaths in conflict states solvable problems.

---

Ebola Is Still Winning the Race

Why it matters: CEPI's 100-day vaccine goal is now part of the UN Political Declaration on Pandemic preparedness. CEPI cited modeling that a 100-day COVID response could have averted 8 million deaths and $14 trillion in losses.

Where things stand: Annie Maxwell, Global Health Lead of Beneficial Deployments, Anthropic, recalled her time at the UN Office of the Special Envoy for Tsunami Recovery. She described a two-hour window in which people could have been evacuated, but no early warning system existed. "We all know that in our line of work, saving time is saving lives," she said.

CEPI's Deputy CEO spoke from the vantage point of Bunia in the Democratic Republic of Congo. The Bundibugyo-strain Ebola outbreak there is the second-largest and the fastest-growing on record. It has infected more than 7,000 people and killed more than 3,600.

Health workers cut test turnaround from about two days to about 100 minutes. Two vaccine candidates, the first ever against Bundibugyo, entered human trials within 11 weeks. COVID vaccines, by comparison, took 326 days. The deputy CEO still called that pace not fast enough.

"The health workers that I met are on their knees with exhaustion, and Bundibugyo Ebola is still outpacing us," said CEPI's Deputy CEO.

Working with Anthropic, CEPI set up a shared partner workspace. It also used Claude to help experts assess vaccine funding proposals faster.

"For my whole career working on outbreak response, it's taken days and days to get an outbreak situation report and understand what's happening on the ground. Now it takes about 12 minutes," the deputy CEO said.

The other side: CEPI warned that the same tools cut both ways. "They can make it easier for bad actors to deliberately create or release harmful biological agents like toxins or viruses," the deputy CEO said of AI and biotech. The deputy CEO argued that a fast response also deters attacks.

What's next: The 100 Days Mission is central to CEPI's new five-year strategy. CEPI has placed biosecurity at the heart of its AI investments since 2023, developed responsible principles for AI in biological design, and plans a pandemic preparedness engine. The UN declaration was under discussion in New York during the week of the event.

---

'Plenty' Aims to Spot Famine Earlier

Why it matters: WFP warned of possible famines in Sudan, South Sudan and Somalia. It said intervening before a crisis costs seven times less than responding after one.

Where things stand: An Anthropic technical staff member spent much of the year with WFP's early warning and forecasting units. The goal was to free analysts from data cleaning and repetitive work.

The result is Plenty, a system that pulls together dozens of sources, including price, conflict and displacement data. It keeps those sources current and produces maps, briefs and datasets.

"Now, Claude is excellent at much of this, but most of it does not need superhuman AI," he said. After five days in Rome, the team demoed new features to country offices preparing El Niño season analyses. WFP collaborators retain responsibility for modeling rigor.

Rania Dagash Kamara, Assistant Executive Director for Partnerships and Innovation, World Food Programme, said WFP feeds about 120 million people in about 80 countries. Seventy percent of its operations are in conflict zones. Last year brought two confirmed concurrent famines, in Sudan and Gaza.

"Now, we don't see AI as a replacement for expertise. We have plenty of that. But it is a force multiplier," Dagash Kamara said.

Decisions and stakeholder impacts: Dagash Kamara said country directors and heads of office, in places such as Al-Fashir and Wajir, validate findings and remain accountable. She stressed data protection, privacy, transparency and human oversight.

She cited a cyclone in the Philippines where anticipatory cash and relocation reached 400 families.

What's next: Anthropic said it will open-source Plenty so the broader humanitarian ecosystem can use it, potentially beyond food insecurity. No release date was given.

---

Capitals Pitch Small Pilots, Big Scale

Why it matters: Governments framed small-market pilots as models that could scale. They named language and local context as the key hurdles to reaching rural populations.

Where things stand: Elizabeth Kelly, Head of Beneficial Deployments, Anthropic, moderated the panel.

Rwanda's Minister of ICT and Innovation cited more than 30 AI use cases across agriculture, health care and education. The minister said Rwanda wants to use its roughly 15 million people as a proof of concept for scale. Anthropic engineers work alongside Rwandan engineers. "We are not going to wait until we have everything before we can make AI useful," the minister said.

Philip Thigo, Special Envoy on Technology, Kenya, said Kenya signed an agreement with Anthropic that day covering health, education, research, creativity and foresight. He said Kenya is also working with Anthropic on languages. "Nairobi is not a rural area, and the rural area is not Nairobi," he warned.

Meenakshi Ramesh, CEO, Pratham Education Fund, said Pratham's work with Anthropic on Teaching at the Right Level can be shared with similar countries. "The grassroots is the laboratory, and then the scale is the government," she said.

Decisions and stakeholder impacts: Looking to 2030, the panelists described who they hope benefits:

• Ramesh hopes policy frees Indian teachers from census and election duties.
• She also hopes AI helps teachers see each student's level. "I firmly believe that the teacher is not removable from the class," she said.
• The Rwanda minister hopes community health workers get decision-support tools with patient history.
• The minister also hopes farmers get timely weather, seed and market information in local languages.
• Thigo hopes for individual agency and choice. He criticized current AI use for emotional advice or companionship.

---

Anthropic's Case: Cheaper Access, Uneven Adoption

Why it matters: Anthropic's own economist said Claude usage tracks national income. That raises the risk that AI's gains flow first to wealthy countries.

Where things stand: Kelly opened the event by calling it an inflection moment that requires continued risk mitigation alongside delivering benefits. She pointed to the essay "Pacing the Frontier" and to the release of Opus 5.5 the same day.

"Price, language, and expertise remain blockers," Kelly said. Her commitments addressed each one:

• A 90% discount on Claude team plans for nonprofits in low- and middle-income countries
• Work to improve model performance in widely spoken Global South languages
• Applied AI and customer success support for nonprofit partners
• Claude Academy AI fluency courses in more languages

She cited Ghana, where Anthropic worked with PlayLab and the Ministry of Education to give 68,000 teachers curriculum-aligned lesson planning tools. "Let's be honest, our industry has had far too much hubris and assumed that technology itself is often the answer," she said.

The other side: Peter McCrory, Chief Economist, Anthropic, warned that early-adopting high-income countries could capture AI's benefits. "If so, this could result in a divergence of living standards, even as AI accelerates global economic growth, not too different from what happened in the late 1800s," he said.

Anthropic interviewed 81,000 people across 159 countries in 70 languages. Optimism was highest, and job-displacement concern lowest, where adoption is lowest. McCrory also highlighted the Economic Futures Research Fund, a $200 million commitment launched earlier this year to fund pilots and randomized controlled trials.

Anthropic's Head of Research for Knowledge Work said the length of tasks AI can complete passed a full working day this spring and is doubling every seven months. He pointed to One Laptop Per Child in Peru, where test scores did not move because curriculum, training and the school day stayed the same.

"Write the benchmarks and show that we're bad at it. You will trick the labs into spending millions of dollars to solve your problem," he said.

---

Minor Items

• Open Evidence, which founder Daniel Nadler said roughly 70% of US doctors use, will deploy free to clinicians in low- and middle-income countries through an Anthropic partnership.
• Anthropic, the Gates Foundation and IDinsight are turning national curricula into knowledge graphs, starting in Nigeria, Rwanda and India.
• Demographic and Health Surveys Program and Yale PopHive data are being connected to Claude via MCP, a standard for linking data sources to AI tools.
• Anthropic cited an AI for Education Alliance partnership in Brazil.

---

Looking Ahead

Watch for action on the UN Political Declaration on Pandemic preparedness, the open-source release of Plenty, and WFP country offices' El Niño season analyses.