The U.S. Senate's Subcommittee on Emerging Threats and Capabilities issued a public challenge to the Pentagon on May 20 that cut to the heart of the largest proposed autonomous-weapons investment in American history: a $54.6 billion budget request for drone warfare is moving faster than the rules governing when those drones can kill. "The policy architecture really has to scale with it," Sen. Joni Ernst (R-Iowa) said at the hearing. "And this is where we probably lag behind."
At the technical center of that spending are two newly unveiled programs from the Defense Advanced Research Projects Agency, or DARPA — one designed to let swarms of drones think and coordinate without any central command node, the other to build robots that sense and react through their physical materials rather than through vulnerable processors and datalinks. Together, they represent DARPA's answer to what Defense One has called the defining unsolved problem of autonomous warfare: how does a small team of human operators effectively command hundreds or thousands of robots when communications may be jammed and the battlefield is moving faster than human reaction time?
DICE: Swarm Intelligence Engineered from the Bottom Up
The Decentralized Artificial Intelligence through Controlled Emergence program, known by the acronym DICE, seeks to build something that has no direct predecessor in military technology: a drone network in which no single node is essential, and where mission-level behavior emerges from local interactions between individual units rather than from commands issued by a central controller.
DARPA's documentation for the program describes the goal as mirroring the scalability principles of the internet, "where robust global behavior emerges from simple, local rules." Individual drones in a DICE-compliant swarm would coordinate with their immediate neighbors through peer-to-peer communication, dynamically forming and reforming into mission subgroups, redistributing roles when units are lost or jammed, and maintaining alignment with commander intent — what the program calls "role coherence" — across multiple operational steps without requiring upward authorization for each decision.
The architecture deliberately addresses the vulnerability of centralized control. A traditional drone network with a command node is one successful radio-frequency jam away from collapse. A DICE swarm, by design, has no such node to disable. The program also explicitly addresses "rogue" agent risk: local inference controls on each drone are intended to constrain emergent behavior so the collective does not develop what DARPA's documentation describes as "misaligned instrumental goals." That concern — that self-organizing systems might pursue unintended objectives — is one of the core technical challenges in any decentralized autonomous architecture. DARPA program manager Susmit Jha is leading the effort, with a Proposers Day scheduled for May 29 in Arlington, Virginia.
Materials Intelligence: Thinking Without a Processor
The companion program, Materials for Physical Compute in Untethered Robotics, aims to eliminate a different bottleneck — the central processor itself. Every robot currently in military service relies on the same architecture: sensors collect data, transmit it to a processor, the processor decides, and actuators execute. That loop introduces latency, power draw, and a communications dependency that adversaries can exploit.
DARPA wants robots whose sensing, computation, and response are encoded in their physical structure — in shape-memory polymers that shift geometry under stress, in piezoelectric materials that convert pressure directly into electrical signals, in pneumatic systems that run local feedback loops without routing information through any central node. The concept has an established name in academic robotics: morphological computation. What DARPA is asking for is a defense-grade, field-deployable version of an architecture that nature has already demonstrated works: the sea anemone contracts when touched not through nervous deliberation, but through the mechanics of its tissue.
The practical payoff in a contested environment is significant. A drone whose responses are embedded in its materials cannot be disrupted by jamming its datalinks — because it has none. At the smallest operational scales, where conventional microprocessors become impractical regardless of communications availability, physical intelligence may be the only viable architecture.
$54.6 Billion and the Doctrine Gap
Both programs are positioned to feed research and ideas to the Pentagon's Defense Autonomous Warfare Group, or DAWG, whose proposed FY2027 budget of $54.6 billion would represent a 243-fold increase over its inaugural FY2026 allocation of $225 million. The figure exceeds the entire Marine Corps budget request for the same year at $52.8 billion, and it sits in the reconciliation portion of the $1.5 trillion FY2027 defense proposal rather than the base discretionary budget. Former CIA director David Petraeus and tech entrepreneur Isaac Flanagan, writing in The Hill, called it "the largest single commitment to autonomous warfare in history."
The risk Petraeus and Flanagan identify is specific and documented. During the past decade of U.S. operations in the Middle East, each Predator drone required nearly 150 support personnel to maintain a single continuous surveillance patrol. The drones were not the limiting factor — the people and organizational structure required to operate them were. The argument they make is that the same failure mode is embedded in the current spending plan: less than two percent of the new investment is directed toward the doctrine, training, and force design needed to deploy autonomous systems effectively. The rest is hardware on order.
Emil Michael, the under secretary of defense for research and engineering and the Pentagon's chief technology officer, acknowledged at the May 20 hearing that autonomous weapons policy "absolutely needs updating" — driven not only by expanded capability but by what adversaries are already fielding and by lessons learned from U.S. operations in Iran. Sen. Elissa Slotkin (D-Mich.) drew a historical parallel at the same hearing: the Roosevelt administration funded the Manhattan Project while concealing its costs from Congress. Today, she said, innovation in autonomous weapons is primarily driven by the private sector, which raises a governance question the government has not yet answered. "I do not believe that a private-sector company should get to decide what the rules are," Slotkin said.
The existing governing document, DoD Directive 3000.09, updated in January 2023, requires that autonomous and semi-autonomous weapon systems be designed to allow "commanders and operators to exercise appropriate levels of human judgment over the use of force." Ernst argued at the hearing that the directive was never designed to contemplate an environment in which AI-driven targeting is integrated with autonomous munitions at the pace now under development.
Minimal Human Oversight: DARPA's Central Design Goal
The operational ambition behind both DARPA programs is explicit: reduce the number of humans required to command a given number of autonomous systems. DAWG's mandate is to enable a small team of operators to direct a large, heterogeneous force across multiple warfighting domains simultaneously — air, land, sea, and potentially space.
The Pentagon moved further in that direction on May 20, selecting Shield AI and its Hivemind autonomy software to integrate swarm capability into the LUCAS drone — a low-cost one-way attack platform the U.S. military reverse-engineered from an Iranian design. Hivemind, unlike traditional autopilot systems, dynamically reroutes mission plans in response to unexpected conditions without requiring real-time human input. Shield AI says a single operator will be able to command an entire swarm operating collaboratively in communications-constrained environments. A demonstration is planned for this fall.
On the organizational side, Marine Gen. Francis Donovan announced the establishment of the SOUTHCOM Autonomous Warfare Command, a new entity within U.S. Southern Command dedicated to deploying autonomous, semi-autonomous, and unmanned platforms across domains. The command will work directly with DAWG to align capability with doctrine across the Western Hemisphere.
First Acknowledged AI-Assisted Civilian Casualty
The doctrine gap is not theoretical. A joint investigation by Airwars and The Independent published in March 2026 identified Abdul-Rahman al-Rawi, a 20-year-old student from al-Qaim, Iraq, as the first acknowledged civilian victim of a U.S. airstrike operation in which the use of AI-assisted targeting was publicly claimed. U.S. Central Command later told investigators it had "no way of knowing" whether artificial intelligence was used in the specific strike that killed him — a contradiction experts described as raising "every red flag." The accountability gap that Ernst and Slotkin described from a Senate hearing room has already produced a documented casualty and an official letter of condolence that declined to explain why it happened.
Whether Congress will enact the full $54.6 billion DAWG request remains unresolved. The figure requires passage through budget reconciliation — a procedural path that bypasses the standard committee authorization process — and Senate Armed Services Committee leadership has already signaled concern about routing a program of this size outside normal oversight channels. But the direction is clear: the Pentagon is committing to autonomous warfare at a scale that would place it alongside nuclear deterrence and carrier strike groups as a primary pillar of American military capability. What it has not yet committed to, senators argued on May 20, is the doctrine needed to use it.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
