OpenAI deploys custom GPTs for disaster relief in Asia

When extreme weather hits, people no longer just wait on emergency hotlines. They open ChatGPT.

During Cyclone Ditwah in Sri Lanka, crisis-related queries on ChatGPT spiked 17-fold. When Cyclone Senyar slammed Thailand in November 2025, message volume jumped 3.2 times. Everyday citizens are already treating AI as an emergency advisor, but frontline responders are still drowning in fragmented data and manual paperwork.

On March 29 in Bangkok, OpenAI hosted its first AI Jam for Disaster Management. Partnering with the Gates Foundation, the Asian Disaster Preparedness Center, and DataKind, OpenAI brought together 50 disaster response leaders from 13 Asian nations to turn LLMs into operational rescue tools.

Building custom GPTs for disaster zones

Participants came from 13 disaster-prone nations across South and Southeast Asia, including India, the Philippines, Indonesia, Bangladesh, Vietnam, and Thailand. They represented government agencies, multilaterals, and frontline NGOs.

Asia bears the brunt of global catastrophes, accounting for roughly 75% of all disaster-affected populations worldwide, according to Humanitarian Action data. The World Bank estimates past disasters have drained over $11 billion from ASEAN economies alone.

Disaster teams usually operate in low-resource setups where stitching together situation reports takes hours of manual work.

Instead of lecturing on abstract AI capabilities, OpenAI mentors worked side by side with officials to build custom GPTs and reusable workflows. The focus was strictly practical: drafting situation reports, conducting rapid needs assessments, and pushing clear crisis communications to the public.

ADPC Executive Director Aslam Perwaiz noted that integrating AI into geospatial risk analytics helps turn raw satellite and Earth observation data into actionable maps and early warnings.

OpenAI confirmed this workshop builds on its OpenAI for Countries initiative launched at Davos, with a second phase of live pilot deployments scheduled in the coming months.

Why it matters

For emergency agencies, the barrier to entry is finally dropping. Responders don't need million-dollar custom models to clear communication bottlenecks. A well-designed workflow on an existing LLM can parse bilingual disaster reports and draft resource allocations in minutes instead of hours.

Strategically, we think this is a brilliant play by OpenAI. Partnering with respected institutions like the Gates Foundation lets OpenAI embed its technology deep into government emergency workflows across the Global South, creating long-term institutional stickiness.

However, crisis response leaves zero room for model hallucinations. A fabricated shelter address or distorted supply estimate could cost lives. Furthermore, severe storms often knock out local power and cellular connectivity, meaning cloud-dependent AI tools must prove they can function when disaster zones go completely dark.