The Pentagon has tapped Shield AI to make its cheapest attack drones fly themselves. The defense-autonomy firm will integrate its Hivemind software as the "AI pilot" for LUCAS — the Low-Cost Uncrewed Combat Attack System, a roughly $35,000 one-way attack munition the U.S. military reverse-engineered from Iran's Shahed-136. An operational demonstration slated for this fall will put a single human operator in command of a swarm of Hivemind-flown LUCAS drones.
From autopilot to AI pilot
Hivemind is Shield AI's flagship autonomy stack, and the LUCAS effort positions it as the decision layer rather than a flight-stabilization add-on. According to the company, Hivemind will let groups of drones "coordinate, maneuver, and adapt together to changing conditions in real time, based on warfighter input," sensing, deciding, and acting "without human intervention" while one operator retains command authority over the formation.
The distinction matters for anyone building autonomous systems: this is intent-level control, not waypoint scripting. Hivemind handles dynamic in-flight rerouting, obstacle avoidance, and real-time threat response on the edge. If mobile air defenses engage and destroy part of a swarm, surviving units reroute and continue the mission. "Hivemind is the AI pilot that makes that mass intelligent," said Brandon Tseng, Shield AI's co-founder and president.
The $35,000 math behind the mass
LUCAS is the U.S. answer to the attritable-drone problem Iran exposed. The roughly 10-foot airframe carries an explosive payload that detonates on impact and costs in the "mid-to-low tens of thousands" — about $35,000 per unit. The Pentagon's chief technology officer said in March the military had produced "just dozens" of the systems, which first launched from a littoral combat ship in December.
The procurement logic is cost asymmetry. Iranian Shaheds proved cheaper to produce than the U.S. interceptors fired to stop them, and the Iran War accelerated demand for low-cost coordinated strike. "It's cheaper to destroy a target, but it's also keeping our war fighters safer," Tseng said, framing autonomous mass as a taxpayer argument as much as a tactical one.
What's confirmed — and what isn't
Shield AI did not disclose a contract value and declined to say whether the selection is tied to an existing award. The deliverable is a fall demonstration in the second half of 2026 in which a single operator directs a swarm of Hivemind-equipped LUCAS drones in coordinated flight.
Why it matters for the autonomy stack
The signal here is software-defined attritable mass: the value migrates from the airframe to the autonomy layer running on it. As $35,000 munitions become the unit of force and one operator scales to a swarm, the differentiator is the edge model coordinating them — and Shield AI is racing to make Hivemind the default pilot before cheap hardware fully commoditizes.
