AI Is Saving Lives in War Zones — Not Just Taking Them. The Root Goes Back to a Red Cross Filing Cabinet.

Key Takeaways

• •SpotlightAI identifies landmines in Ukraine in under 0.2 seconds per drone image — 23,000+ threats found across 5,338 hectares
• •Tarjimly served 609,000 refugees in 2024 with 61,000 volunteer translators, valued at $4.7 million in free services
• •WFP's AI gives 60-day advance warning of food crises across 90+ countries
• •ICRC's AI matching engine reunited 7,000 families in 2024 alone
• •VIEWS conflict prediction system correctly identified 7 of 10 deadliest conflict countries in 2024

Timeline

In February 2025, an AI system called SpotlightAI processed its one millionth drone image over a Ukrainian minefield. It took 0.2 seconds per image. It had already flagged 23,000 explosive threats buried across 5,338 hectares of farmland, roads, and neighborhoods. Nobody was hurt in the process.

That same month, the Peace Research Institute Oslo published a machine-learning forecast showing Sudan's projected death toll for 2026 had more than doubled in a single month's update. The model had correctly predicted 7 of the 10 deadliest conflict countries in 2024. It was sounding an alarm that human analysts were still debating.

These are not the AI stories that make headlines. There are no autonomous weapons here, no killer drones, no dystopian scenarios. This is AI doing something quieter and arguably more important: saving lives in the places where life is cheapest.

## The Minefield Problem

A single drone image analyzed in 0.2 seconds. Over 23,000 explosive threats identified. In a country where 138,500 square kilometers are contaminated — an area the size of Florida — every fraction of a second matters.

Ukraine is the most mine-contaminated country on Earth. Roughly 138,500 square kilometers — an area the size of Florida — are riddled with landmines, cluster munitions, and unexploded ordnance from more than two years of war. Traditional demining is painstaking: a human with a metal detector clears about 20 square meters per day. At that rate, clearing Ukraine would take centuries.

Safe Pro AI, a Virginia-based company, built SpotlightAI to change the math. Drones fly over contaminated areas capturing high-resolution imagery. The AI processes each image in under 0.2 seconds, identifying surface-visible mines and ordnance that human eyes would miss at altitude. By early 2025, the system had analyzed over one million drone images and identified more than 23,000 explosive remnants of war.

In June 2024, the HALO Trust — the world's largest humanitarian landmine clearance organization — partnered with Amazon Web Services in a $4 million initiative. AWS provides the cloud infrastructure and AI tools; HALO has flown 542 drone flights over Ukrainian minefields, generating 11 terabytes of imagery. Their AI is learning to identify explosive remnants in drone footage and to analyze satellite imagery for damaged buildings and human activity near minefields, so clearance teams can prioritize where civilians are most at risk.

This isn't replacing human deminers. It's telling them where to look first. When you have an area the size of Florida to clear and people are dying every week, knowing where to start is everything.

## Feeding People Before the Crisis Hits

284,000 missing persons documented worldwide — up 68% since 2019. Behind every number is a family that doesn't know if someone they love is alive.

The World Food Programme feeds roughly 150 million people each year. In war zones — Sudan, Gaza, eastern Congo, Myanmar — getting food to the right place at the right time is a logistics problem of staggering complexity. Roads are destroyed. Supply chains are severed. Prices spike overnight. Borders close without warning.

In 2024, WFP deployed an AI tool called SCOUT that optimizes procurement decisions — what to buy, where to source it, when to ship, and how to route delivery. In its first year, SCOUT saved $2 million on sorghum sourcing in West Africa alone. WFP projects $25 million in annual savings at full scale.

But the more remarkable tool is PREDICT, built with the Danish Refugee Council. It uses machine learning to anticipate displacement patterns up to four months in advance, analyzing open-source data on conflict intensity, food prices, weather, and historical population movements. WFP's broader machine-learning systems now provide 60-day advance warning of food security crises across more than 90 countries.

The difference between knowing a famine is coming in 60 days versus learning about it when people are already starving is not incremental. It is the difference between pre-positioning supplies and conducting an emergency airlift. Between $2 per meal and $20 per meal. Between lives saved and lives counted.

## The Translation Gap

When a Syrian family arrives at a refugee camp in Jordan, or a Sudanese mother reaches a border crossing in Chad, the first barrier is almost never physical. It's linguistic. They can't explain their medical symptoms. They can't understand the asylum process. They can't read the forms that determine whether they'll be allowed to stay.

Tarjimly was founded in 2017 to close this gap. The app connects refugees and asylum seekers with volunteer translators in real time — phone calls, video calls, text translation. By 2024, the platform had 61,000 volunteer translators, many of them immigrants and refugees themselves, and served 609,000 refugees. That year the service facilitated over 1 million minutes of interpretation and nearly 1.4 million words of translation, valued at $4.7 million in services — all free.

In 2024, Google.org gave Tarjimly a $1.3 million grant to build "First Pass," an AI tool that generates an instant initial translation for human volunteers to refine. The system is fine-tuning large language models for underrepresented languages — Rohingya, Swahili, Tigrinya — languages that major AI systems handle poorly because there simply isn't enough training data.

Every new conflict creates a surge. Afghanistan 2021. Ukraine 2022. Sudan 2023. Gaza 2024. Each time, Tarjimly's servers spike with desperate people who need someone — anyone — who speaks their language.

## Finding the Missing

Here is the number that should stop you: 284,000 missing persons documented worldwide as of December 2024. That's a 68% increase since 2019. Behind every entry is a family that doesn't know whether someone they love is alive or dead, imprisoned or free, in a mass grave or a hospital ward.

The International Committee of the Red Cross has been searching for missing people since 1870, when it began tracing wounded soldiers in the Franco-Prussian War. By the end of World War I, its International Prisoners of War Agency had compiled 7 million index cards. The Geneva Conventions of 1949 formalized the right of families to know the fate of their relatives — a right that exists on paper but is agonizingly difficult to enforce.

Today, the ICRC's Missing Persons Digital Matching Project uses AI to cross-reference multiple databases — from NGOs, governments, international organizations — and flag potential matches for human investigators. The AI uses predictive analytics to detect patterns and connections within and between cases. Visual recognition technologies analyze photographs of documents, military ID tags, and written reports recovered from war zones.

In 2024, the ICRC located 16,000 people and reunited 7,000 with their families. The scale of the problem dwarfs the solution, but the direction is clear: AI doesn't replace the painstaking human work of investigation, but it finds needles in haystacks that no human team could search manually.

## Predicting Where War Will Kill Next

The Violence & Impacts Early-Warning System (VIEWS), developed by the Peace Research Institute Oslo and Uppsala University, uses machine learning to produce monthly forecasts of armed conflict fatalities up to three years ahead. The system generates global country-level predictions and sub-national forecasts at fine geographic resolution for Africa and the Middle East.

In 2024, VIEWS correctly identified 7 of the 10 deadliest conflict countries. Its forecasts flagged Sudan's rapidly deteriorating situation months before the humanitarian crisis reached its worst point. These aren't vague warnings — they're quantified predictions with confidence intervals, updated monthly, available to any humanitarian organization planning where to deploy resources.

The Armed Conflict Location & Event Data Project (ACLED) produces a complementary Conflict Index and annual Watchlist, providing data-driven analysis of where violence is intensifying. Together, these systems give aid organizations something they've never had before: a credible forecast of where the worst suffering will happen next.

## Seeing Destruction from Space

When bombs fall, the first thing that disappears is accurate information. How many buildings were destroyed? Where are civilians still sheltering? Which roads are passable? In the chaos of active conflict, these questions can take weeks to answer on the ground.

The UN Satellite Centre (UNOSAT) uses AI-assisted analysis of satellite imagery to produce comprehensive damage assessments. In Gaza, as of October 2025, UNOSAT documented approximately 198,273 damaged structures — roughly 81% of all buildings in the territory. The assessment categorizes damage levels: 123,464 destroyed, 17,116 severely damaged, 33,857 moderately damaged, and 23,836 possibly damaged.

In 2025, researchers published an open-source tool in Nature Communications using Sentinel-1 SAR satellite time series to map war destruction at building level. The system runs on Google Earth Engine and can generate updated damage estimates every 12 days, validated across Beirut, Mariupol, and Gaza. It's open-source — meaning any researcher, journalist, or humanitarian organization can use it.

This matters for accountability. When there are disputes about what happened, satellite imagery analyzed by AI provides evidence that is difficult to deny. It matters for reconstruction planning. And it matters right now, in active conflicts, for directing aid to the neighborhoods that need it most.

## Healing Where There Are No Doctors

Syria lost nearly half its doctors between 2010 and 2018 — from 0.529 to 0.291 per 1,000 people. In conflict zones across the world, healthcare infrastructure is often the first casualty. Hospitals are bombed. Medical staff flee. Supply chains for drugs and equipment collapse.

In Jordan's Zaatari refugee camp, AI-driven health assistants help Syrian refugees navigate available healthcare services — guidance on vaccinations, maternal health, and chronic disease management. AI chatbots provide health information in refugees' native languages, bridging the gap when human translators aren't available.

The Omdena + Colour the World Foundation partnership built an AI chatbot for PTSD risk assessment in war and refugee zones. Health workers using the tool reduced evaluation time by 40% and improved diagnostic accuracy by up to 30%. In settings where a single mental health professional might serve thousands of displaced people, that efficiency isn't a luxury — it's triage.

AI-powered drones are delivering vaccines and emergency medical supplies to areas that vehicles can't reach — destroyed roads, active combat zones, flooded terrain. The World Food Programme and Médecins Sans Frontières have both deployed these systems.

## The Root

The through-line from an ICRC volunteer writing names on index cards in 1918 to a machine-learning system cross-referencing databases across 50 countries in 2024 is not technology. It's the same stubborn human insistence that every person matters — even in the places where the world has looked away.

AI doesn't end wars. It doesn't stop the bombs. It doesn't negotiate ceasefires or hold war criminals accountable. What it does, when pointed at the right problems by the right people, is make the difference between a family that waits forever and a family that finds each other. Between a farmer who steps on a mine and a farmer who doesn't. Between a refugee who can explain her symptoms and one who suffers in silence.

The technology is not neutral — nothing ever is. But in the hands of the ICRC, WFP, HALO Trust, Tarjimly, and dozens of other organizations operating in the world's worst places, AI is doing something that matters more than any benchmark or stock price: it is making the machinery of compassion faster, smarter, and more precise.

The headlines will keep screaming about killer robots. Meanwhile, in a minefield outside Kherson, a drone camera clicks and an algorithm thinks for 0.2 seconds. Safe. Not safe. Safe. Not safe. Move the children this way.

(Sources: Safe Pro AI, HALO Trust, AWS, WFP, ICRC, Tarjimly, UNOSAT, VIEWS/PRIO, Uppsala University, Google.org, Nature Communications, Frontiers in AI, Omdena)

How did this make you feel?

Recommended Gear

Keep Reading

AI for Good

Amazon Cut 16,000 Jobs Citing AI — The Luddites Fought This Same Battle in 1811

6 min read

AI for Good

Yann LeCun Left Meta to Build a Brain — The Theory Behind It Came From a 1943 Book

6 min read

AI for Good

AI Is Now Detecting Wildfires Within 10 Minutes of Ignition. The Root of Fire Lookouts Goes Back to the Roman Empire.

7 min read