The Pentagon's New Autonomous Warfare Command Is Already Redefining 'Human Control'

On September 30, Defense Secretary Pete Hegseth stood before roughly 600 officers and enlisted personnel at Marine Corps Base Quantico and announced that the Department of War would build the Autonomous Warfare Command, or AUTOWARCOM — a four-star combatant command dedicated to scaling drones, robotics and AI across the armed forces. It would be the military's twelfth combatant command and the first new one stood up since U.S. Space Command in 2019. Hegseth signed a same-day memo, 'Project Agincourt: Pathway to the Autonomous Warfare Command,' directing the Pentagon's unmanned-systems portfolio manager to deliver an implementation plan within 30 days and to begin working with Congress on the legislation needed to formally create the command by its target date of October 1, 2027.

Hegseth framed the move in sweeping terms, saying the department would 'institutionalize' and 'supercharge' what he called the era of 'super intelligence, robotics, and autonomous systems,' pointing to the drone-saturated Russia-Ukraine war as evidence that the character of combat is changing faster than military bureaucracy can keep up.

From Announcement to Doctrine

Two weeks later, at a panel during the Association of the U.S. Army's annual meeting in Washington on October 14, Hegseth and the service secretaries filled in more of the picture — and signaled the command is still taking shape. 'We're still iterating, we're still adapting, we're still learning,' Hegseth said, adding that AUTOWARCOM would govern training and doctrine for unmanned systems, not just acquisition. The Army had announced the day before that it would convert its Futures and Concepts Command into a new Futures and Autonomy Command; the Navy and Air Force have not yet announced their own component structures, though the Air Force is already testing autonomous fighter aircraft at Creech Air Force Base in Nevada.

The more consequential statement came from Army Chief Technology Officer Alex Miller, who said the service wants 'appropriate levels of human judgment' built into autonomous warfare — but that for fast-reaction missions like counter-battery fire and air and missile defense, humans should be 'on the loop' rather than making every individual firing decision. His reasoning: the volume of drones and munitions involved in modern warfare has outgrown what people can process in real time.

Why the Wording Matters

That distinction — 'on the loop' versus 'in the loop' — is not a technicality. It is the exact fault line that arms-control researchers and AI-safety advocates have been pointing to for years as the place where accountability for a lethal decision can quietly shift from a person to a system. 'In the loop' generally means a human must approve a strike before it happens; 'on the loop' means a human supervises and can intervene, but the system can act on its own in the interim. For a single munition that distinction may be minor. For a command designed to coordinate swarms of autonomous systems at machine speed, critics argue it is the mechanism by which oversight erodes without anyone deciding, in a single moment, to remove it.

The replacement of troops with machines makes the decision to go to war easier.

That line comes from Amnesty International, which has renewed its call for a binding international ban on lethal autonomous weapons in response to the Pentagon's announcement. The group's broader argument is twofold: that autonomous systems lower the political and human cost of initiating conflict, and that weapons built to operate without constant human direction are also easier to lose control of — through battlefield capture, illegal transfer to third parties, or simple malfunction at a scale no single operator is watching.

A Different Kind of Catastrophic Risk

Most of the AI catastrophic-risk conversation this year has centered on chatbots and coding agents acting outside their intended scope inside corporate and government networks. AUTOWARCOM is a reminder that the loss-of-control question also has a kinetic version, where the systems in question are armed. None of the officials who spoke at the October 14 AUSA panel addressed safety testing, failure rates, or formal human-control policy for the command in any detail — the public discussion so far has been driven by acquisition and doctrine officials, not by anyone describing how the Pentagon will verify that 'appropriate levels of human judgment' hold up once autonomous systems are fielded at scale.

AUTOWARCOM still needs congressional authorization to formally stand up, and the Pentagon's own timeline gives it roughly a year to work out exactly what human oversight will mean in practice across counter-battery fire, missile defense, and the swarms of drones the department is racing to field. Whether that year includes rigorous, independent testing of the autonomy itself — rather than just the hardware around it — is the question safety researchers outside the Pentagon are now watching most closely.