Public transcript

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

By: Peter H. Diamandis
Length: 36 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Palmer Luckey sits down with Peter H. Diamandis and guests at an optimism summit to discuss Anduril's wildfire X Prize win, a new secret $10 million X Prize on animal intelligence, and his views on AI risk and autonomous weapons. He argues autonomous weapons are ancient, that defense autonomy must live forward in the field, and that Anduril won't build humanoid robots because defense demands specialized, overwhelming performance. He closes with lessons from running an online community as a teenager and advice on what AI does and doesn't change about being a founder.

Key points

Palmer Luckey and Anduril won the $2.2 million wildfire X Prize, including $1 million from Lockheed, for autonomous fire detection and response technology.
Luckey is funding a new secret $10 million X Prize focused on understanding biological intelligence and communicating with wild animals in their own language.
He is not worried about AI doom; he is worried about evil people using moderately competent AI to do evil things.
Autonomous weapons have existed for thousands of years, from pitfall traps and landmines to the Aegis system's fully autonomous mode, and new technology is just making them better.
Autonomy must live forward in the field, not at headquarters, because jammed, intercepted, or hacked communications would disable tens of thousands of assets.
Anduril builds its own specialized, deterministic, auditable models for radar and sonar data rather than using consumer foundation models.
Anduril will not build humanoid robots or generalist defense robots because defense demands overwhelming, specialized performance, not shared economies of scale.
Anduril succeeded in defense by being a product company using its own money, displacing systems that cost needlessly ten times what they should.
The new Meta VR glasses are praised for off-boarded compute and batteries, high-quality pass-through, and being a third the price and an eighth the weight of Apple's product.
Running ModRetro, an online community he started at 14 or 15, taught Luckey organizational lessons he carried into Oculus and Anduril.

Questions it answers

00:05Wildfire X Prize win

What did Anduril win and how long has this technology been possible?

Anduril won the $2.2 million wildfire X Prize, including $1 million from Lockheed. Luckey notes Anduril built an autonomous firefighting tank from 2017 to 2019, and the technology has been possible for over a decade.

$2.2 million prize, $1 million from Lockheed
Anduril competes with Lockheed's fighter jet division but partners on payloads
Autonomous firefighting has been possible for over a decade

01:31Autonomous firefighting pipeline

How does Anduril's autonomous firefighting system work?

The system detects fires, classifies them, rejects false fires like cigarettes and road flares, matches the fire type to the right cost-effective asset, and hits the target. Unlike military targets, fire response time drastically changes which tools you need.

Pipeline: detection, classification, rejection, matchmaking, hitting the target
False fires like cigarettes and campfires must be rejected
A fire metastasizing demands the right tool before it grows too large

03:32New $10M animal intelligence X Prize

What is the new X Prize Luckey is funding?

A secret $10 million prize about understanding biological intelligence, specifically how animals think, aiming to communicate 10 times better than the state of the art with wild animals in their own language.

$10 million prize, details secret for a month
Focus on animal intelligence, not pushing AI forward
Could yield more power- and time-efficient AI

05:30P-doom and AI risk

What is Palmer Luckey's P-doom?

He is not worried about AI doing the killing; he is very worried about evil people using moderately competent AI to do evil things. He thinks AI will massively reduce costs across society, like agriculture and textiles already have.

AI doom is way down his list of threats
He criticizes recent converts to doomerism who never studied the threats
AI will bring economies of scale to every part of society

07:43Edge autonomy vs. centralized control

Why should autonomy live in the field rather than at headquarters?

Because relying on a central compute location requires high-bandwidth command and control to thousands of assets, which enemies can jam, intercept, or hack, or they can just attack the control center. Underwater and radio-silent operations also demand local autonomy.

Jamming or hacking communications disables centralized fleets
He cites the Trade Federation's droids falling over in Star Wars
Radio emissions expose you; underwater you cannot communicate

11:44Autonomous weapons are ancient

Why does Luckey say autonomous weapons are ancient?

Defined as weapons executing programmed intent independent of the operator, they go back thousands of years: pitfall traps, poison traps, landmines, sea mines, torpedoes, Vietnam-era missiles, and the Aegis system's fully autonomous mode. New technology just makes them better.

Aegis on destroyers can shoot down anything in range without human confirmation
SeaWiz and SeaRAM have fully autonomous modes
The decision of where to send a weapon remains with people

13:44Anduril's specialized AI models

Will Anduril build its own foundation models?

Anduril has built its own models since 2017. Consumer foundation models are largely non-deterministic and non-auditable, and Anduril needs to process raw radar and sonar data, not text in and text out.

Anduril is a decade beyond the foundation model question
Specialized defense models have few consumer applications
LLMs make sense for some defense uses but not missile guidance

14:50No humanoid robots or generalist killer robot

Why won't Anduril build humanoid robots or a generalist robot?

No. In defense you must massively overpower adversaries, so a 5% performance edge is worth a lot, and the loss from lack of specialization makes generalists non-viable. He wants Anduril to build things that wouldn't exist otherwise, though he expects humanoids to be a great interface automating legacy manned systems.

Humanoids make sense for manufacturing, hospitality, and medicine
Defense is like motorsports: exotic, specialized, continuously pushing limits
Anduril also skips quadcopter attack drones and space launch

Notes

Wildfire X Prize

Anduril won the $2.2 million wildfire X Prize, with $1 million of that from Lockheed.
Anduril built an autonomous firefighting tank starting in 2017, finished in 2019; the technology has been possible for over a decade.
The system handles detection, classification, rejection of false fires (cigarettes, campfires, road flares), matchmaking fires to the right assets, and hitting the target.
With fires, unlike bombs, the difference between a 10-minute and an hour response requires totally different tools.

New $10M X Prize

Luckey is funding a secret $10 million X Prize around understanding biological intelligence, especially how animals think.
Goal: communicate 10 times better than the state of the art with the smartest species, including wild animals in their own language.
He believes animal intelligence could inform more power- and time-efficient AI.

AI Risk and Autonomy

His P-doom is low; he worries about evil people using moderately competent AI, not AI itself.
Autonomy must live forward in the field: centralized compute requires high-bandwidth command and control that can be jammed, intercepted, or hacked (his Star Wars Trade Federation example).
Autonomous weapons are ancient: pitfall traps, poison traps, landmines, sea mines, torpedoes, Vietnam-era missiles, and the Aegis system's fully autonomous mode.

Anduril's Approach

Anduril has built its own models since 2017; consumer foundation models are non-deterministic and non-auditable, unsuitable for radar and sonar processing.
No humanoid robots: defense requires overwhelming, specialized performance, not shared form factors; Anduril builds things that wouldn't exist otherwise.
Anduril is a product company using its own money, displacing systems that cost needlessly 10 times what they should.

Odds and Ends

John Carmack's 1%-chance AGI argument inspired Luckey's bet on defense.
He criticizes billionaires who do nothing, calling credibility an underutilized resource.
He praises the new Meta VR glasses: off-boarded compute and battery, Transformo Fabric (his CV1 invention), a third the price and an eighth the weight of Apple's product.
Running ModRetro from age 14-15 taught him the "two-body problem" of managing brilliant people who hate each other.
AI doesn't change the founder's job much: founders were always displaced from hands-on tech; if you want to work on technology, work for somebody else.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.