Two hikers in their early 20s spent the evening of June 23, 2026 lost off the Dead Horse Gap track in Kosciuszko National Park, Australia, and a Fire and Rescue NSW drone running AI detection software found them within five hours of the 7pm report. The agency says it was the first operational find for its new system, and the inspector who oversaw the search believes it cut days off the operation.
Five Hours From Evening Report to Midnight Walkout #
The two men failed to reach a rendezvous point on the Dead Horse Gap walking track, about 22 miles (35 km) southwest of Jindabyne in the Snowy Mountains, and were reported overdue at about 7pm local time on Tuesday, June 23, according to the Guardian’s June 27 report.
NSW Police stood up an alpine search and rescue team with State Emergency Service volunteers. As darkness fell and the cold set in, police asked Fire and Rescue NSW for support, and firefighters from Jindabyne launched a remotely piloted aircraft from the search command post.
The drone’s AI software picked out the pair in bushland about 1,640 feet (500 m) off the walking track. The thermal camera then zoomed in to confirm the find, according to the agency’s June 25 release.
The hikers helped their own cause. In the darkness, they waved the red light of a mobile phone at the aircraft, and the operator answered by switching on the drone’s spotlight to mark their position for ground crews, who also used the onboard speaker to talk to the pair.
Finding them was the fast part. Canberra outlet Region Media reported that ground teams still needed about four hours to reach the men, who were guided to safety around midnight, suffering mild exposure and declining medical treatment.
None of that terrain is casual in an Australian winter. The NSW National Parks service describes the track as sitting almost 2,000 feet (600 m) above the valley floor, and Thredbo’s own trail guide recommends walking it between November and April, not late June.
The Algorithm Calls a 35 Percent Chance of Being Human #
The AI detection system works as a triage layer: it scans the drone’s camera feeds in real time, assigns each object it sees a probability of being a person, and cues the pilot, who then flies closer and confirms the hit with the thermal camera.
Inspector Phillip Eberle, Fire and Rescue NSW’s regional duty commander for the alpine area, walked Region Media through the logic. “The AI scans the area and scans the information that it can see on its camera and says, ‘OK, this item that I’m looking at has a 35 per cent chance of being human’,” he said.
The threshold sits low on purpose. “If we’d mark that as being, say, 100 per cent [sure that it was a human], it wouldn’t pick it up,” Eberle explained. The system tolerates false alarms rather than risk scoring a real person too cautiously.
The software flags a candidate, and the humans close the loop. “It’ll tell the pilot, ‘This is a person’, and then we’ll fly over and double-check that it is,” Eberle said.
The Guardian reported that the aircraft carries four cameras, including one with infrared imaging, and that the software can analyze thousands of images in real time while telling people apart from animals and other heat sources. Neither the agency’s release nor the coverage names the drone model or the software vendor.
First Operational AI Find for the NSW Drone Team #
Fire and Rescue NSW confirmed this was the first time its drone team’s AI detection system located missing people in a live operation, and the agency credits the technology with turning a search that could have run for days into a single evening’s work.
“It’s definitely helped make what could have been a long-term incident into a very short-term incident,” Eberle told the Guardian. The technology meant “less time on the mountain and less time looking for people, and reduced risk so that the volunteers can get home to their families quicker.”
He also framed the stakes in plain terms: “What could have happened in 12 hours or 24 hours, had we not found them?”
Acting Superintendent John Marzol used the agency’s release to urge hikers in the snowfields to notify the National Parks and Wildlife Service before heading out. Standard advice, except this rescue turned it into a live demonstration.
The pattern is familiar to DroneXL readers. In April 2025, I covered North Shore Rescue in Vancouver pairing thermal drones with AI that analyzes every pixel of the feed, and in August 2025, Italian rescuers found a hiker missing for 10 months after AI software combed through 2,600 drone frames.
DroneXL’s Take #
I’ll be direct, no algorithm pulled anyone off that mountain. The AI watched camera feeds faster and more patiently than human eyes can, then flagged a shape with a modest probability of being a person. A pilot flew over and made the call.
That division of labor is exactly why this worked. Eberle’s 35 percent line is the most useful detail in the story: a search team wants a paranoid algorithm. Tune the threshold low and let the pilot sort out the false alarms, because the miss you can’t afford is the hiker the software scored too cautiously.
The Italian Alps case from August 2025 ran the same play: AI flagged the anomalies and humans reviewed every candidate. Same architecture, different mountain.
The caveat worth keeping: these two men were awake and signaling with a red phone light. Whether the system performs as well on a subject who can’t signal, unconscious or hypothermic, is an open question this rescue doesn’t answer.
None of the coverage I’ve read names the aircraft or the software vendor, and I want both on the record, along with detection stats: hit rates, false positive counts, and time from launch to first flag. Agencies that publish those numbers make a stronger case for AI search and rescue than any press release does.
Two men walked off a winter mountain around midnight instead of spending days on it. That’s the standard this technology now has to keep matching.