As a historic drought desiccates the West and fire season rages year-round, officials from California to Colorado are relying on camera networks equipped with artificial intelligence to detect flames long before humans can.
Early detection in California is “becoming one of the clearest climate‑era advantages,” said Phillip SeLegue, Cal Fire’s staff chief of intelligence. “The system allows firefighters precious minutes in a landscape where fires grow faster every year.”
Cal Fire, the state’s firefighting agency, uses a system called Alert California. It’s a network of more than 1,200 cameras installed in key areas from the redwood forests in the North, all the way down to the southern border with Mexico.
With sweeping 360-degree views and a boost from AI, the system sends fire managers an alert at the first sign of danger. This week, atop a mountain in Southern California, a camera captured the earliest moments of the Sandy Fire as it erupted Monday outside of Simi Valley.
In 2025, the cameras detected 915 fires before the members of the public reported them, SeLegue said.
A for-profit model
Alert California was developed by researchers at the University of California, San Diego. But AI-powered firefighting networks have also created new opportunities for technology companies.
Sonia Kastner, CEO and co-founder of San Francisco-based Pano AI, said the company has cameras perched on cell towers and mountaintops monitoring 50-million acres across 17 U.S. states, along with parts of Australia and Canada.
“We never would have guessed we’d be sitting here in 2026 doing deployments in Georgia, in the Dakotas, in New Jersey,” said Kastner, whose company launched in 2020. “Wildfire risk is spreading east at an alarming rate as conditions get worse and worse and worse.”
As the risk grows, demand for the technology grows along with it.
Arizona Public Service, the largest utility in Arizona, expects to have 71 AI cameras installed by the end of summer. Getting a closer look at a fire helps the utility decide if it should turn off its powerlines in the event of a disaster, said Cindy Kobold, APS supervisor for fire science and meteorology.
“It’s a game-changer,” she said from a command center in Phoenix, where video screens glow with weather data and live feeds from cameras that have already been installed.
Mike Alexander, who runs the Douglas County Office of Emergency Management in Colorado, said Pano AI cameras alerted his team last month to the Bear Creek Fire burning in rugged terrain southwest of Denver. No humans were able to see it, but the cameras could. Pano’s technology monitors about 20% of the entire state, according to the company.
“Early detection is huge,” said Alexander, whose crews were able to pinpoint the fire and put it out. “It makes the difference between catching a fire and not catching a fire.”
AI can’t fix some fire problems
AI systems of all kinds are imperfect. In the context of fire fighting, a column of dust can be confused with smoke.
“In the early days, there were some funny occasions where we detected what we thought was smoke, but it was actually a snow blower on the mountain in Colorado,” said Kastner of Pano AI.
With every recorded image, new data helps refine the algorithm and improve its accuracy. Still, Kastner said a human reviews every alert before it’s sent out because, “if you’re in the middle of fire season battling active blazes, the last thing you want is false alarms.”
AI also cannot fix the foundational issues that have led to catastrophic wildfires in the United States, said Arizona State University’s Hannah Kerner.
Under the Trump administration, the federal government has fired workers and continues to fall behind on its goal of reducing fuels on public land.
“We know that invasive grasses in Maui are a huge problem,” said Kerner, who uses AI and satellite data to predict fire risk. “It’s important to remember that there are on-the-ground solutions that can also help to reduce fires, not just letting those continue while we develop systems to detect them better.”
Meanwhile, as companies profit from the technology, UCSD professor Falko Kuester warned that troves of data collected from networks of AI-powered cameras should remain in the public domain.
Kuester, a co-principle investigator for the Alert California network, said the state’s cameras record 10 million images every day and provide a critical lifeline to first responders.
“It shouldn’t be paywalled away, firewalled, locked away,” said Kuester. “In this case the shareholder is really everyone.”
This article was originally published on WBUR.org.
Copyright 2026 WBUR