In the early hours after a powerful earthquake struck Venezuela on June 24, 2026, first responders knew that people needed help. They did not know who needed it first. Carlos Rivas, a disaster risk management specialist at Catholic Relief Services, had already confirmed that his staff and volunteers survived and could work. Then he confronted the questions filling his inbox: "Where are the hotspots? Where should we go?"
Getting that wrong is costly. A team can spend twelve hours crossing broken terrain toward one town, acting on incomplete reports, only to discover that a community in the opposite direction was hit far harder. Roads and bridges may be gone. In the first hours of a disaster, every hour spent gathering information is an hour not spent delivering aid.
Rivas's answer arrived by the next morning: a damage map produced with HASTE — High-speed Assessment and Satellite Tracking for Emergencies, a tool from Microsoft's AI for Good team. Microsoft says HASTE and its humanitarian partners have used the system after 23 major disasters in 16 countries since 2023, including wildfires in France, floods in Libya, an earthquake in Türkiye and a cyclone in Madagascar.
The workflow is deliberately simple. Analysts first establish a baseline from high-resolution satellite imagery showing buildings as they were. After the event, they label a handful of examples — this building is damaged, that one is not — to show the model what to look for, and HASTE applies those patterns across a much larger territory. The tool is free and open-source, and Microsoft says it needs only a few hours of training on a specific situation before it can assess areas that would otherwise take people days or weeks to comb through by hand.
That human step is the point, not a caveat. HASTE is not a model that knows what a collapsed roof looks like in the abstract; it is a pipeline that lets a human teach it the visual signature of a particular disaster, then scales that judgment across thousands of images. Caleb Robinson, a principal research scientist for AI for Good, notes that every disaster leaves a different footprint and that buildings vary dramatically from region to region — which is why the labeled examples matter more than any general capability.
The headline number is the compression of time. Nora Lindstrom, who leads global digital programming at Catholic Relief Services from Helsinki, says an assessment like the one HASTE produced overnight would have taken her group roughly two weeks through a manual process of analysts and imagery. Two weeks is far too long to matter in the immediate aftermath of an earthquake.
But speed alone is not the goal. "If you have fast garbage, that's actually worse," says Kevin White, a senior director at Microsoft's AI for Good Lab. "AI allows you to get high precision, fast — and that's the holy grail." Rivas's version is plainer: "It's about going fast and having certainty about your decision."
The clearest account of why certainty matters comes from the United Nations. Fawad Hussain Syed, a humanitarian affairs officer with the UN Office for the Coordination of Humanitarian Affairs in Geneva, says post-disaster information usually arrives in fragments — messages, news reports, social-media posts and scattered field observations that are hard to verify. "You're saving lives by providing this data to the team who is doing search and rescue," he says, because it gets rescue teams to the right places, with dogs and machinery, in time to pull living people out of the rubble. Depending on the disaster, ten to fifteen international teams may land at an airport and ask the same first question: where do we go? "This is exactly what this data then provides," Syed says.
OCHA did not consume the Venezuela analysis in isolation. The surge of imagery and damage assessments from many providers began "flooding the system," says Zachary Arno, a data scientist at OCHA, so he built a web viewer hosted on Azure that brought HASTE's assessment together with eight other disaster-mapping products in one place. Microsoft's data arrived with detailed metadata and confidence rankings, Arno says, which made it easier to fold into a broader picture. Leonardo Milano, OCHA's data science team lead, frames the payoff simply: "It's the combination of speed and scale that is really the added value."
The technology is most useful when damage is uneven — some communities flattened while neighbours are barely touched. That is precisely the pattern a uniform keyword or single-pass survey tends to miss, and precisely where an analyst-trained model can direct finite resources to the hardest-hit places.
Response is only the beginning. In August 2026, as wildfires burned through the Spokane, Washington, area, Tristan Allen of the state's Emergency Management Division asked Microsoft what was possible for counting structures inside the fire line. Firefighters were still fighting, and formal assessments would come later, but an early read on the scale of loss let officials begin estimating how many displaced households they would need to support. "If you're looking at thousands of structures lost, you're going to surge more resources earlier," Allen says — instead of waiting for the slow, guess-based version of the same process.
Because HASTE is open-source, organizations adapt it to their own needs. Rivas extended the analysis to evaluate routes, flag obstacles and suggest alternatives, so aid workers would not drive into impassable terrain. He also built a map of tarp distribution to show donors how their money sheltered people in the worst-hit neighbourhoods. Catholic Relief Services is drawing on damage data from Hurricane Melissa, which struck Jamaica in October 2025, to plan shelters in areas vulnerable to future storms — turning a response tool into a preparedness one. And Joseph Porcelli, who works with public agencies at Nextdoor, is connecting the platform with HASTE data so local governments can target specific geographic areas during future emergencies.
The benefit is not limited to disasters that make global headlines. Rivas points to recent fires in Bolivia and floods in Peru, Paraguay and Ecuador as cases where smaller events, ignored by international coverage, still gain a faster read from AI-assisted analysis. By August, he and Syed say, the process had grown faster still: HASTE maps were informing decisions within an hour, not a day, of a Colombia earthquake.
What is emerging, in Rivas's words, is "a fast-evolving culture of consuming this data." The significance is less about any single map than about the loop it creates: responders learn to trust and act on satellite-derived evidence, which pressures the tools to become faster and more accurate, which in turn produces more trust. Microsoft's White says the goal is to give first responders the most accurate, up-to-date information as quickly as possible — and to encourage organizations to download the tools and build the capability themselves rather than depend on a vendor.
There are honest limits. HASTE still needs a human to train it for each event, its accuracy benchmarks are not published, and it solves a narrow problem — spotting destroyed buildings from orbit — rather than the whole of disaster response. Distribution, medical care and shelter remain human work. But that narrowness is the lesson. Applied AI earns its place not by being general or glamorous, but by taking a task that once consumed two weeks of skilled analyst time and turning it into an overnight map that changes where the trucks are sent.
Sources
Microsoft (Source) — "When disaster strikes, AI helps responders reach people more quickly," by Susanna Ray, published September 22, 2026: https://news.microsoft.com/source/features/ai/when-disaster-strikes-ai-helps-responders-reach-people-more-quickly/
Reported details, figures and quotations are drawn from that Microsoft article describing HASTE, part of Microsoft's AI for Good program.