Texas A&M develops AI to sort disaster posts, but it has not reached live emergency use

For now, Texas A&M researchers are developing something narrower: a model that sorts crisis-related social media posts into useful categories. In a November 6 update, the university said the system remains under development and has not been deployed during an active disaster.

The update describes work associated with CrisisSense-LLM, a study dated May 26, 2026. The work is led by Bo Li, Ali Mostafavi and Kai Yin in Texas A&M’s Zachry Department of Civil and Environmental Engineering.

Turning a stream of posts into categories

The project targets the information overload that accompanies emergencies such as hurricanes. Officials must follow weather conditions, damage reports, requests for assistance and search-and-rescue needs while making quick decisions. The researchers’ approach uses a large language model to organize crisis-related social media information, with the aim of making relevant reports easier to locate.

Its classification scheme covers three broad areas: disaster-related, weather-related and humanitarian information. The model also labels posts by event type, informativeness and humanitarian information type. That gives the research more than one sorting task: it considers both the subject of a post and additional categories intended to make the information useful to emergency personnel.

The study used a benchmark of human-annotated social media posts drawn from multiple disaster events. Those human labels provide the reference for the classification task. The university’s update does not give accuracy figures or measured time savings, so it offers no numerical basis for judging how reliably or quickly the system would help responders.

Ultimately, we want to move from classification toward a more interactive AI system that can better support situational awareness during disasters.

Bo Li, Texas A&M postdoctoral researcher

Conversation is the planned next step

Li said the team plans to build an AI agent that emergency professionals can talk to when disaster strikes.

The National Science Foundation funded the research. The university identifies the underlying study as “CrisisSense-LLM: instruction fine-tuned large language model for multi-label social media text classification in disaster informatics.” Its current research task is organizing posts; the team’s stated ambition is to reduce time spent sorting reports so agencies can focus sooner on decisions for affected communities.