AI emergency alerts are gaining traction as officials look for faster, clearer ways to reach the public during disasters, but experts warn that accuracy, authenticity, and trust remain critical hurdles.
Earlier this month, after Hurricane Isaias came ashore on the Gulf Coast as a Category 3 storm, emergency management specialist Laura Myers received a call from an Alabama meteorologist. Myers, who leads the University of Alabama’s Center for Advanced Public Safety, was told a local emergency manager had posted a graphic suggesting the Alabama coast could see a “sting jet” of intense winds. The catch, according to Myers, was that the image had been created by an AI chatbot and did not reflect sound meteorology.
“That county emergency manager was trying to get his stakeholders to focus on the risk in that county, and so he had asked AI to help him,” Myers said. “It did what he wanted it to do, but it was not meteorologically accurate.”
Most emergency agencies still prohibit staff from using generative AI, yet Myers and other specialists say the tools are already being used informally across state and local departments. Vendors are pitching hundreds of AI products for everything from flood damage assessments to resource coordination, and many officials and startups are particularly focused on public messaging.
One motivation is simple. Many emergency managers struggle to write rapid, plain-language alerts under pressure. Jargon, missing context, and extraneous details can blur the most important information during fast-moving crises.
“A lot of agencies are really limited in resources and the number of people that they have,” said Jeannette Sutton, a disaster researcher at the University at Albany who built a comprehensive dashboard with alert templates. “They recognize it as a priority, but they don’t prioritize it until they realize they should have. They’re just trying to get their job done without screwing something up.”
Emergency managers coordinate disaster response and recovery at the local and state level, forming the first line of defense after extreme weather. Many say “trust is our currency,” and they earn or lose it in the way they communicate with residents.
Communication failures have compounded emergencies before. In 2016, two neighboring California counties issued “voluntary” and “mandatory” evacuation orders for the same wildfire, creating confusion. In Australia in 2019, officials used the phrase “Watch and Act” to warn of wildfires, a term a later government review said led to “significant confusion.”
Sometimes the stakes are life or death. After the 2017 Thomas Fire in California, heavy rains triggered a devastating debris flow in Montecito. Santa Barbara County struggled to convey risk to residents still traumatized by the fire and experiencing what Sutton calls “warning fatigue.” Officials pushed alerts to everyone in a wildfire “voluntary evacuation zone,” telling people to stay alert and be ready to leave. Many residents interpreted the message as optional. The disaster killed 23 people and injured more than 160.
Sutton said poor warnings erode credibility. “They’re losing the opportunity to be that voice to help people to be safe.”
How AI emergency alerts could help
On paper, AI looks well-suited to assist with alerts. FEMA trains officials on best practices, but in a real emergency many do not recall every step. A large language model can sift through vast numbers of past alerts, draw on research about what works, and help refine a draft in seconds.
“If you’ve only sent one [alert] in your life, and that was 15 years ago, and you need to remember all your training, and you’re thinking ‘Hey, there’s a chemical plume coming my way, the phone’s ringing, the radio’s blaring, I need help,’ the AI is probably gonna do a better job than you are,” said Rob Dale, a former meteorologist who leads emergency management in Ingham County, Michigan.
Dale has experimented with AI to draft practice messages but has not sent one publicly. He said he would use a chatbot to draft incident communications if needed, while keeping human control over what gets published.
Some developers argue the tools are ready now. EM1, a startup building AI for emergency managers, says it trained a model using Sutton’s research and that the system meets or exceeds human performance on many benchmarks in scenarios such as wildfires and tsunamis.
To demonstrate, Sutton shared a real alert from a New Mexico county about rising water on the San Juan River:
The original alert
“Breeched banks. The water is rising. EVACUATE NOW. McGee Park open for people and pets, RV.”
She asked EM1’s model to critique and revise it. Within seconds, it produced a more complete version:
The new alert
“San Juan County: Floodwater is rising near [location]. People in [evacuation area]: LEAVE NOW for higher ground. Do not walk or drive through floodwater. McGee Park is open for people and pets. Check [official update source] for updates.”
The model addressed key gaps by identifying the sender, clarifying locations and who must evacuate, offering safety guidance, correcting a misspelling of “breached,” and pointing to an official source for updates.
EM1 says it works with about 100 government agencies and school districts, though none are using it to fully automate weather alerts yet. For now, agencies are testing AI for planning and internal preparations. Even so, experts believe many managers already lean on chatbots to draft messages. Sutton said trainees often present polished alerts they generated in seconds using widely available tools.
“Official adoption is pretty low, but people are unofficially using it, absolutely,” said Sarah Miller, a professor of emergency management at Pierce College in Washington and longtime instructor at the National Disaster Preparedness Training Center. “They’re using their own personal accounts, or they’re using free ones.”
Trust and warning fatigue with AI emergency alerts
The calculus is not simple. Faster, clearer AI emergency alerts could save lives, but the tools could also weaken public confidence. Residents wary of AI or already skeptical of government may reject posts or graphics flagged as AI-generated. In Syracuse, New York, a public employees union sued the city over its use of an AI service to handle non-emergency 911 calls.
“We trust things that we perceive are authentic and care about us,” said Ginny Katz, founder of HazAdapt, which advises officials on technology use. “AI-generated content does not feel authentic.” More frequent, rapid-fire alerts could also intensify warning fatigue.
As with other AI deployments, outcomes depend on the user. A seasoned manager might safely accelerate communications with AI support. A novice under pressure may not be able to assess an AI draft or catch errors from models such as ChatGPT or Claude.
Despite these concerns, many short-staffed departments continue to experiment. It remains unclear how the spread of AI-generated public messaging will affect trust in government over time.
“If you had a human sending you a text message versus a computer sending you a text message, which one would you believe is being more accurate?” asked Justin Snair, founder of Preppr, another company offering AI tools for emergency managers.
The answer, he said, depends on the person receiving it.
