“We do not have a decade to transition our procurement and doctrine,” Defense Secretary Pete Hegseth wrote in a Sept. 30 memo that set the Pentagon on a focused push to field autonomous and robotic systems faster than past cycles allowed.
The mandate: AUTOWARCOM and an urgent timeline
Hegseth’s memo directs the creation of an Autonomous Warfare Command, “slated to launch by next October,” with a mission to promote “not a particular drone, but the ability to adapt in combat” against “realistic opposition.” The language frames the problem as one of speed and adaptability: on modern battlefields, responses that once took months or years now must happen in weeks.
Start with the mission — not the robot
The memo and recent practitioners interviewed in the source material push a single organizing idea: ask commanders and operators what they need, then design autonomy to meet that mission. The Maven Smart System is held up as a model. The system, described as an AI intelligence platform “widely adopted by the U.S. military and NATO forces,” benefited from a 2024 CNAS report’s finding that “a clear, urgent, but broad problem definition permitted Maven to more directly meet user needs.”
Gen. Frank Donovan, commander of U.S. Southern Command and the former head of the Defense Autonomous Warfare Group, told audiences in May that commanders most often want domain awareness and data they can use immediately: “Most of the systems we're looking at are primarily our domain awareness systems, and…We can match the robots to the environment whether it swims, it flies, it has feet, whatever it does.” Donovan emphasized that partners on the ground — those who “have the access and placement where they live” — need robots to deliver information from places where boots are not present.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleDesign choices: separate brains and bodies, standardize, and modularize
A former senior defense official with direct experience fielding AI software urged a separation of autonomy software from vehicle platforms. Drawing on analogies to the auto industry, the official noted that commercial self-driving businesses separated vehicle manufacture from autonomy stacks: “the automakers build great vehicles and partner with autonomy companies for the driver.” By contrast, the official argued, Pentagon practice too often asks manufacturers to build autonomous versions of existing remote-controlled platforms.
The same official recommended modular autonomy packages and open standards so software can be upgraded as threats change. That view was reinforced by a recent NATO exercise: in September, the alliance’s Centre for Maritime Research and Experimentation (CMRE) completed its largest annual Robotic Experimentation and Prototyping with Maritime Unmanned Systems exercise and demonstrated an architecture in which “algorithms [were] looking at what the mission was looking like and automatically giving tasks to drones,” João Alves, a principal scientist at CMRE, reported. CMRE’s result followed a policy shift among partner militaries to adopt a common standard for purchased systems — not a new sensor or algorithm, but a procurement-level change.
Trust and safety: the Patriot lessons
Past autonomy failures underscore the human and procedural challenges ahead. A 2017 study of autonomy safety failures involving Patriot missile batteries concluded that the central difficulty is sustaining “high reliability in a complex and unpredictable operational setting.” The study, by Army engineering psychologist John Hawley, highlights a dramatic case: the 2003 downing of an F-18 by a Patriot in Iraq, traced to a tactical director’s order to “Bring your launchers to ready,” which was interpreted as an engagement directive even though that was not the director’s intent.
Hawley’s assessment and subsequent commentary in the source material stress that even with mission clarity, standards, and shared data, commanders must still solve the problem of when to rely on autonomous systems. The same concern has become more acute as AI agents have in other contexts executed actions beyond user intent; the source warns that “human trust can’t be automated.”
What this means for commanders, Pentagon buyers, and NATO partners
- Commanders and operators: Expect emphasis on tailored data services and domain-awareness tools rather than platform-centric demonstrations; training and clear intent-setting will be critical to build reliable human-machine team decisions.
- Pentagon buyers and procurement leaders: The memo and practitioners push a buying model that separates autonomy software from platforms, specifies modularity and open standards, and prioritizes policy shifts that make interoperability possible across services and suppliers.
- NATO and partner militaries: The CMRE experience shows that agreement on standards — a procurement and policy choice — can be as decisive as new hardware or algorithms, enabling algorithms to allocate tasks across heterogeneous drones.
If AUTOWARCOM is to chart a lasting path, its leaders will need to combine those procurement and engineering lessons with rigorous training and clear operational doctrine. The source material closes on a pointed reminder: past successes and failures must guide the command, and leaders must acknowledge that although software can assign tasks and manage data, “human trust can’t be automated.”
Original story: The Pentagon’s autonomy command should draw on lessons old and new (Defense One)
