The 2025 wildfires in Southern California were the worst on record since the 2018 Camp Fire that destroyed the town of Paradise. The combined effect of the Palisades and Eaton fires in Los Angeles was the destruction of over 16,000 structures and over 34,000 acres of land. Thirty people were killed, and more than 150,000 were displaced from their homes. Across the region, communities both within and proximate to the burn zones now face serious health challenges from exacerbation of respiratory, cardiovascular, and neurological illnesses through exposure to wildfire smoke, to losses of jobs and livelihoods leading to increases in mental health issues and food security strains.
Although the impact of each of these events was uniquely devastating to the communities that they touched, wildfires have become a ubiquitous reality in the lives of Californians. Across each of these events, key questions for emergency responders are the same: Have people evacuated? If so, where are they going? Are they safe? And which local governments have primary responsibility for those recently displaced?
The California Governor’s Office of Emergency Services, or Cal OES, looks at multiple sources for the answers, including unique public–private partnerships. One of these partnerships includes information from private technology firms like Meta, which maintains a vibrant and long-standing Artificial Intelligence for Good program. This real-time, aggregate data on evacuee movement was available almost immediately during the Palisades and Eaton fires to assist emergency managers with situational awareness.
The maps provided by Direct Relief’s CrisisReady partnership with Harvard Data Science Initiative and AI for Good at Meta helped Cal OES to confirm that the majority of evacuated Californians were headed to neighborhoods just outside of Altadena, including South Pasadena and Glendale, as well as north towards Thousand Oaks and Burbank, and to areas along the southern coast as well.
As the frequency and severity of climate disasters increase, the need for this type of effective data, and increasingly, artificial intelligence, is becoming more urgent than ever. The same types of collaborative partnerships that have defined the approach in California towards the open integration of novel data insights about human mobility into emergency response also have much to tell us about a future where different forms of AI can be applied safely and effectively for community resilience. And yet, across the United States, there is skepticism about the adoption of AI without sufficient attention paid to the life-saving interventions that these novel tools can facilitate.
In a day-long workshop in December 2024 at Meta’s offices in Fremont, California, a group of county and state level emergency managers came together with Direct Relief, CrisisReady, and Meta to reflect on past events and explore the advancing edge of AI and novel data sources for effective response to disasters, and in June, a group across California state emergency management and health services met in Los Angeles to discuss how the recovery from this year’s fires could be hastened by leveraging AI.
During these workshops, California state employees discussed how multimodal speech and text translation models are opening up new possibilities for rapid community information and diverse public engagement. Similarly, participants explored how generalized image segmentation is improving the ability to use aerial and satellite images for wildfire detection and containment, as well as damage assessment. These partners also described how large language models like Llama are enabling first responders to summarize and absorb unprecedented amounts of mission-critical data and report on crisis events.
Only a few weeks after that event, the wildfires erupted across Los Angeles. Maps were quickly produced by the Direct Relief / CrisisReady team and shared with Cal OES and others to show early evidence of evacuation patterns from Altadena and Palisades into neighborhoods near the respective burn zones, as well as a number of other communities across Southern California.
While those maps in themselves were not determinative for the response, they provided fast and significant insights about the rate and direction of population flow from the most affected areas, and improved visibility into emerging needs and aid dynamics, which played out over the coming months.
Other AI-produced maps based on satellite imagery and building footprint data showed damage to infrastructure within days of the fires starting. Social connectedness maps, showing density of connections on the Facebook platform between the Palisades and Altadena neighborhoods with surrounding areas, improved predictions of the drivers and magnets of displacement.
The collaboration networks formed over years of engagement were put to the test in rapid succession throughout the events in Los Angeles, but there is still much to learn. In July 2025, a larger group of technologists, public sector agencies and healthcare leaders gathered in downtown Los Angeles at the headquarters of the Community Clinic Association of Los Angeles County to reflect upon the events of the past six months, the future of disaster resilience in the city, and the types of emerging data and tools that disaster managers, health providers, and humanitarians can now use to bolster their efforts.
The conclusion of those conversations was that communities and agencies need to be closely involved in technology adoption and use so that sustainable trust can be established and insights can be shared and acted upon effectively.
“In responding to disasters, it is important that our leaders have every potential tool available to make the most informed and most viable decisions. Partnerships with organizations like Direct Relief bring additional visibility to a disaster’s changing landscape, allowing those in charge to have a better picture of what is happening on the ground,” said Abby Browning, Chief, Office of Private Sector / NGO Coordination, for Cal OES.
As the United States heads into a new era of AI innovation, what is mostly clear is that models need to be open and accessible wherever possible in order to ensure public confidence. Openness at the model level also needs to be collaborative wherever possible so that governments can transparently and responsibly adopt new technologies that don’t remain behind paywalls or which may be developed through opaque, bespoke agreements. In an environment of openness, early adopters in the nonprofit sector can also play a translational role between tech companies and government without needing to serve as vendors, and instead by focusing on shared innovation agendas and social goals.
The future may be arriving fast, but the best version of it for safer and more resilient communities has been – and will continue to be – open and collaborative partnerships. And in the fight against wildfires, there’s no time to debate whether new technology is worth deploying at all – we should leverage every tool available to hasten the speed of detection and containment of these events for the protection of communities everywhere.
