Self-Driving Tech Built an Attack Helicopter

Anduril and Archer just turned the air-taxi supply chain into a fully autonomous attack helicopter—and Washington is still arguing about who gets to download Kimi K3.

What's new: On July 20 at the Farnborough Airshow, Anduril and Archer Aviation (NYSE: ACHR) unveiled Thunder, a Group 5 autonomous attack rotorcraft designed to fly beside crewed Apaches and the Army's future Cheyenne fleet as a "loyal wingman."

Why it matters: The near-surface fight—the low-altitude corridor where attack helicopters operate—has become one of the most lethal zones on the battlefield. Cheap drones, loitering munitions, and persistent surveillance have pinned crewed rotorcraft at escalating risk. Thunder is Anduril's bet that AI formation flight can put mass back on the U.S. side without putting more pilots in the kill box.

The hardware:

• Series hybrid-electric tiltrotor built on a dual-use platform co-developed with Archer
• Payload capacity Anduril calls "Apache equivalent"—up to 10 Hellfire missiles, 16 air-launched effects, or 76 70mm rockets
• Counter-UAS munitions in the nose to protect both the drone formation and the crewed aircraft it escorts
• Packs into a standard 40-foot ISO container for transport by land, sea, air, or rail

The autonomy: Thunder runs on Anduril's Lattice for Mission Autonomy software, which manages formation flight, separation from crewed aircraft, and mission execution without pilot control of each drone. Anduril President Chris Brose told reporters the design draws on three years of development and lessons from the company's YFQ-44A Fury aircraft, now in production for the Air Force's Collaborative Combat Aircraft program.

The self-driving crossover: The hard problem wasn't the airframe—it was flying fast and low without hitting terrain or colliding with a manned helicopter beside you. Shane Arnott, Anduril's SVP of programs and engineering, said the team "spent the last couple of years working with technologies that are coming out of the self-driving car industry" to solve that. The autonomy stack has been flying on surrogate aircraft for two years. First flight of Thunder itself is planned for 2027, with production targeted for 2029–2030.

The Ukraine lesson: Brose said Thunder's engineering is "really based on the lessons that the United States and our European allies are learning in Ukraine." He envisions three to six Thunders flying with a single Apache or Cheyenne—multiplying sensors and weapons while keeping crews farther from the most lethal threats.

The commercial subsidy: Thunder is the defense variant of a clean-sheet VTOL platform Archer and Anduril built together, leveraging Archer's electric-propulsion expertise and optimum-speed tiltrotor technology. Archer CEO Adam Goldstein called it "the most sophisticated vertical lift aircraft ever made." Archer plans to announce commercial customers for the civilian variant later this week—the classic dual-use playbook where defense budgets fund what passenger eVTOL could not.

The procurement context: The unveiling lands months after the Army killed its Future Attack and Reconnaissance Aircraft program and began searching for a CCA-style wingman for rotorcraft. Anduril was also down-selected for the British Army's Project NYX, which seeks loyal wingman drones for Apache helicopters. Brose acknowledged the Army aviation modernization track record of "failed starts" and "false starts"—then pitched Thunder as the system that finally breaks the pattern.

The governance gap: This is lethal autonomy shipping on a commercial-aircraft timeline while Washington's AI policy fight stays fixated on chatbots. President Trump's June executive order and the new Gold Eagle clearinghouse govern frontier language-model access and cyber vulnerability patching—not autonomous weapons that carry Hellfires. The Future of Life Institute's July safety index gave the top lab a C+ while noting several pivoted toward defense contracts.

What Anduril won't say: Executives declined to name a price. Brose called Thunder "attritable"—cheap enough to lose—but acknowledged that "attritability is obviously in the eye of the beholder."

Convina's view: Silicon Valley spent July repricing trillion-dollar language-model IPOs while Anduril quietly industrialized the deadliest application of agentic AI yet—a self-driving stack that flies beside Apaches, carries ten missiles, and fits in a shipping container. The policy class is still drafting guest lists for GPT-5.6. The battlefield is already fielding loyal wingmen. If Washington wants to lead on AI safety, it needs a rulebook for what just landed at Farnborough—not another clearinghouse for software patches.