28 September 2026
Heard in AI

Anduril's wildfire XPRIZE win went beyond spotting fires

Anduril won $2.2 million in XPRIZE's wildfire contest, including a $1 million detection bonus, after controlled tests in Alaska. Founder Palmer Luckey says the system must also reject decoys and send equipment that can reach a fire in time.

A briefing reports one development at a point in time. We may correct or clarify it later; a new development gets a new briefing. How our formats work

Based on Moonshots with Peter Diamandis, episode published 28 September 2026 (recorded 25 September 2026)

Anduril Industries has won $2.2 million in XPRIZE's wildfire competition. Palmer Luckey, the company's founder, says spotting a fire was only the first step. "You also need to reject false fires," he told Peter Diamandis, the founder of XPRIZE. Then the system must pick a response that can reach the fire before it gets out of hand.

Diamandis congratulated Luckey on stage at Moonshots Live 2026. The conversation was recorded on September 25 and published as an episode of Moonshots with Peter Diamandis three days later. Luckey also founded Oculus VR, the virtual-reality company. Anduril builds AI-powered defense systems.

What the prize asked for

XPRIZE announced the winners on September 23. The $11 million competition started in 2023 and had two separate tracks. One was for detecting fires from space. The other, which Anduril won, was for autonomous response, meaning systems that act without a person steering them. Teams had to detect a high-risk fire as it started, anywhere in an area of 1,000 square kilometres, and put it out completely within ten minutes. Nearby decoy fires had to be left alone.

Anduril took first place and $1.2 million. It also won a separate $1 million bonus for detection and intelligence, sponsored by Lockheed Martin. Together these make up the $2.2 million Diamandis cited on stage, "a million of that from Lockheed," as he put it. XPRIZE said Anduril stood out for linking every step, from detecting a fire to putting it out. Dryad took second place and $800,000; XPRIZE recognised its network of 170 sensors and its drones for working autonomously from start to finish. AURA Foresight came third and won $500,000. It was recognised for meeting the ten-minute detection requirement.

Luckey noted that Anduril and Lockheed are both partners and rivals. In defense, he said, Anduril competes with Lockheed's fighter-jet division and works with the company on some of its payloads.

Luckey's five steps

During a one-minute video of the team's work at the test site, Luckey described the job in steps. First the system detects fires and sorts them by type. The step he said viewers could not see was ruling out false alarms: "who's the guy smoking a cigarette? … what's a campfire? What's a road flare?" Only something that "shows the characteristics of real fire" should be reported to people and to emergency services.

Next comes what he called matchmaking: pairing each fire with the right equipment in the area. "You want to basically use the most cost effective thing that can respond and get there in time," he said. That choice depends on the fire. A fire that is growing faster may need a different response, and so may one in woods rather than on a brushy hillside. He summed up the sequence as "detection, classification, rejection, matchmaking, and then going and hitting the target."

Why minutes change the equipment

Luckey said Anduril's military systems go through many of the same steps. The difference is how much delay matters. In most military work, he said, reaching a target in an hour instead of 10 minutes "usually doesn't make as much of a difference." With fire, he said, the gap between a 10-minute and an hour-long response means "totally different tools." In his words, "a bomb is a bomb whether or not it takes a day or an hour or a minute to get there. But when a fire is metastasizing, you need to pick the right thing that can put it out before it gets too large."

What happened in Alaska

The results came from controlled tests, not firefighting in real service. XPRIZE's log of the final round says three finalist teams were tested from June near Nenana, Alaska, with support from the University of Alaska Fairbanks. Each test included eight hours of monitoring and up to three small fires set on purpose. Judges scored speed, accuracy, autonomy and how well each fire was put out.

The log also describes Anduril's setup. Sentry towers watched the ground using infrared sensors, which detect heat, and computer vision, which is software that interprets camera images. Vertical-takeoff aircraft from other companies covered ground the towers could not see. Anduril's software, called Lattice Fire, combined and checked these observations. It then sent out the nearest Ghost-X aircraft. That aircraft used its own infrared sensor to pinpoint the fire and dropped balls of fire retardant on it. Meanwhile, the other sensors kept watching the fire and looked for new ones.

An old idea, a new urgency

Luckey said he and Diamandis had "been trying to get this autonomous firefighting stuff going for a long time." He said one of Anduril's first two products was a self-operating firefighting tank, which it started building in 2017 and finished in 2019. He also remembered the prize being announced in 2019, but XPRIZE says the competition launched in 2023.

Diamandis called it "insane that we let fires become conflagrations before we put them out." Luckey pushed back on people saying the technology has only now made this possible. In his view, it "has been possible for over a decade." What changed, he said, is that it has "become politically impossible to ignore the problem now."

XPRIZE says a further phase of field trials and work toward operational adoption comes next.

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02

Connected ideas and articles

From the conversation

Podcast episodes

Moonshots with Peter Diamandis

Palmer Luckey: Autonomous Weapons Are Ancient and Why Anduril Won't Build Humanoids | Moonshots Live | EP #295

Episode published (recorded )This article draws on 0:12–0:40, 0:44–0:51, 0:52–1:01 and 1:01–3:29 (approximate times)

Article history

Updates to this article

Tags