Emergency Managers Turn to AI for Alerts After Faulty Hurricane Graphic Sparks Concern
A county emergency manager’s AI‑generated storm map proved inaccurate, highlighting the growing but risky use of generative AI in public safety communications.
AI is rapidly infiltrating emergency management, but its accuracy and impact on public trust remain uncertain.
AI‑Generated Graphic Triggers Alarm
Earlier this month, Hurricane Isaias struck the Gulf Coast as a Category 3 storm. Laura Myers, director of the University of Alabama’s Center for Advanced Public Safety, received a call from an Alabama meteorologist about a local emergency manager’s graphic that showed a “sting jet” of high winds threatening the Alabama coast.
The graphic had been produced by an artificial‑intelligence chatbot at the request of the county emergency manager. While the AI delivered the requested visual, Myers said it was “not meteorologically accurate.” The manager’s intent was to focus stakeholder attention on county‑specific risk, but the result misrepresented the storm’s behavior.
Generative AI Use Is Widespread Despite Restrictions
Most emergency management agencies officially prohibit employees from using generative AI tools. Nevertheless, Myers and other experts believe the technology is already pervasive in state and local disaster departments. Hundreds of companies market AI applications for flood‑damage assessment, resource coordination, and especially public communication.
“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 of Albany. She has built a dashboard with templates for disaster alerts, noting that agencies often recognize the need for better communication only after a failure occurs.
Challenges in Human‑Written Alerts
Emergency managers frequently struggle with public alerts. Human alerts can contain jargon, omit critical details assuming public knowledge, or include irrelevant information that distracts recipients. Sutton cited examples of flawed communication: in 2016, adjacent California counties issued conflicting evacuation orders for the same wildfire, and in 2019 Australia’s “Watch and Act” phrase caused “significant confusion.”
In Montecito, California, after the 2017 Thomas Fire, a heavy rainstorm triggered a deadly mudslide. Santa Barbara County sent alerts to the “voluntary evacuation zone” that asked residents to “stay alert… and be prepared to leave immediately.” Many residents interpreted the notice as optional, contributing to 23 deaths and over 160 injuries.
Potential Benefits of AI‑Assisted Alerts
Proponents argue that alerts are an ideal use case for generative AI. FEMA offers training on best practices for alert composition, but officials may forget details under pressure. A large language model can scan thousands of past alerts, synthesize research on effective messaging, and edit a draft in seconds.
Rob Dale, former meteorologist and current emergency‑management director in Ingham County, Michigan, said, “If you’ve only sent one [alert] in your life, and that was 15 years ago… the AI is probably gonna do a better job than you are.” Dale has used AI to draft practice alerts but has not yet needed to send a live alert. He would retain final control over any public release.
Balancing Trust and Technology
Emergency managers view trust as our currency. Their communication directly builds or erodes public confidence. Sutton warned that poor warnings lose the opportunity to be that voice to help people to be safe. While AI could help craft clearer messages, the technology’s current inaccuracies—exemplified by the faulty “sting jet” graphic—underscore the need for rigorous oversight.
As AI tools become more embedded in disaster response, the balance between speed, accuracy, and public trust will define their ultimate value.
