Defense contractor Anduril is testing something straight out of science fiction - military drones controlled by large language models similar to ChatGPT. At a classified Texas military base, the company demonstrated autonomous fighter jets that can receive voice commands, coordinate attacks, and eliminate targets with minimal human oversight. This marks a dramatic shift in how AI is being weaponized for modern warfare.
The scene unfolding at a secret military base 50 miles from the Mexican border reads like a techno-thriller, but it's very real. Anduril, the defense contractor founded by Palmer Luckey, just proved that large language models can control swarms of killer drones with chilling efficiency.
During a classified demonstration, four prototype fighter jets codenamed "Mustang" appeared on the horizon over the Texas desert. When a simulated Chinese J-20 stealth fighter appeared on radar screens, a simple voice command - "Mustang intercept" - set everything in motion. An AI model similar to the one powering ChatGPT parsed the order, coordinated with the drones, and responded in a calm female voice: "Mustang collapsing." Within minutes, the autonomous aircraft had converged on their target and destroyed it with virtual missiles.
This isn't just another military tech demo. It represents a fundamental shift in how the Pentagon thinks about AI warfare. Anduril is developing a full-scale autonomous fighter called Fury through the Air Force's Collaborative Combat Aircraft program, designed to fly alongside human pilots as an AI wingman. The company calls it "Sergeant Chatbot at your service."
The timing couldn't be more significant. Federal AI contract funding exploded by 1,200% between August 2022 and August 2023, according to a Brookings Institution report. The Department of Defense drove most of that spending surge. Now, President Trump's administration is doubling down with the first-ever dedicated AI allocation in the defense budget - $13.4 billion for AI and autonomy in 2026.
That massive funding shift has Silicon Valley's biggest players scrambling for their piece of the military AI pie. This year alone, OpenAI, Google, Anthropic, and xAI each secured military contracts worth up to $200 million. It's a dramatic reversal from 2018, when Google famously pulled out of Project Maven over employee objections to military AI development.
"The ambition that is a bit scary is that AI is so smart that it can prevent war or just fight and win it," Georgetown University AI researcher Emelia Probasco told WIRED. "Like some sort of magical fairy dust." She's not alone in that concern. Current LLMs remain too unreliable and unpredictable for direct control of lethal weapons systems.
But that hasn't stopped the arms race. Anduril and Meta recently partnered on a $159 million Army contract to develop AI-powered augmented reality helmet displays for soldiers. The system promises to deliver mission-critical information while using next-generation AI models to interpret battlefield conditions in real-time.
The Ukraine war accelerated this push toward autonomous warfare. Low-cost drones equipped with computer vision have proven devastatingly effective on both sides. American strategists now believe whoever controls advanced AI technology will dominate future conflicts - which explains the aggressive efforts to restrict China's access to cutting-edge chips and models.
Former Navy fighter pilot Michael Stewart, who spearheaded AI experiments with the Fifth Fleet, sees fully autonomous soldiers as inevitable. "In 10, 15, or 20 years, you're going to have robots that are pretty autonomous," Stewart told WIRED. "That's where you're going."
The implications are staggering. Unlike today's drones that require human operators, LLM-controlled weapons could make kill decisions independently while explaining their reasoning in plain English afterward. We're not just automating warfare - we're creating machines that can justify their actions in human terms.
Palantir now runs the evolved Project Maven system, called Maven Smart Systems, which has become one of the military's most widely deployed AI tools. The platform excels at intelligence gathering by parsing massive datasets and analyzing aerial imagery. LLMs also show promise in cyber warfare, given their ability to write and analyze code at superhuman speeds.
But serious questions remain about reliability and accountability. Current AI models still hallucinate, make unpredictable errors, and operate as "black boxes" that even their creators don't fully understand. Putting such systems in control of lethal weapons represents an unprecedented leap into uncharted ethical territory.
The Pentagon seems willing to take that leap. Defense contractors are making bold promises about AI capabilities, backed by unprecedented funding and political support. The race is on to deploy these systems before adversaries do the same.
The successful demonstration of LLM-controlled fighter drones marks a watershed moment in military AI development. With $13.4 billion in Pentagon funding and major tech companies now fully committed to defense contracts, autonomous weapons systems are rapidly moving from science fiction to battlefield reality. But as these AI systems grow more sophisticated, the fundamental questions about accountability, reliability, and the ethics of machine-controlled warfare become more urgent. The technology is advancing faster than our ability to understand its implications - and that gap could prove more dangerous than any enemy weapon.
