Six Thousand Choices an Hour

Back in 2023, Ukrainian forces lost an estimated ten thousand drones each month across the front lines—an average of 300 every day. The Royal United Services Institute reported that Russian electronic warfare caused most of those losses by jamming satellite signals and severing pilot controls.

A listener wrote in to ask whether future reconnaissance drones are becoming autonomous digital commanders, picking targets and ordering strikes while soldiers merely watch. That question hits home because machine-speed warfare raises the urgent concern of who holds the trigger when the machine moves at light speed. Ahead, we trace what shattered remote piloting, how onboard microchips take over from vulnerable radio links, and why rules of engagement still require a person to hold the brake.

The short answer is that drones are not turning into sovereign commanders. Rather, they function as edge-computing tactical nodes. Instead of beaming uninterrupted raw video back to distant servers, onboard processors run calculations right where the data is gathered, slashing latency. The aircraft acts like a rapid matching engine, classifying threats and passing along alerts, but it functions entirely under human supervision.

This shift began because remote links proved too fragile. Analysts trace the vulnerability back to 2011, when an American RQ-170, a reconnaissance drone, fell into Iranian hands through GPS spoofing. That weakness severely disrupted radar networks and communications during the opening two days of the 2022 invasion. Russian jamming blinded radars and severed teleoperation, where a remote pilot steers an aircraft with continuous joystick commands over the air. With spotting crippled, planners recognized they were stuck in a DDIL environment—one where communications are denied, degraded, intermittent, or severely limited. If jamming cuts that lifeline, how can a flying sensor guide artillery and infantry when its link to the outside world is almost completely silenced?

To survive in that silence, an aircraft has to stop broadcasting and do its thinking in the air. Relying on those onboard calculations means the platform no longer depends on distant cloud servers to make sense of what it sees. Christopher Jenks points out that executing perception and navigation onboard reduces reliance on remote processing for time-critical functions. As Michael Clegg of Supermicro puts it: "By processing incoming data at the edge, less information needs to be sent to the cloud and back. This also significantly reduces processing latency."

The sheer size of video shows why the old way broke down. A continuous live feed can chew through 11 gigabytes just idling over empty ground waiting for something to happen. Sorting that information onboard slashes transmissions by 20 to 1. Instead of flooding the airwaves, the drone squeezes out lightweight event packages—essentially 50-kilobyte digital postcards that slip through narrow cracks in enemy jamming.

Building those compact updates requires onboard sensor fusion, where the aircraft combines optical cameras, thermal imagers, and radar into one unified tactical picture. Machine-learning models sift through that incoming collection like an automatic filter, discarding the miles of blank terrain, water, and empty sky. As Jenks observes, troops rarely need the full footage; what they need are clear target tracks and priority alerts.

Stripping away that clutter changes how fast troops can fight. Brett Daniel pointed out that data-transmission bottlenecks have long kept operators from acting quickly. Ukraine's GIS Arta software, a battlefield system that distributes fire requests, showed what happens when targeting coordinates flow instantly across dispersed artillery units. In analysis by Major Nickolas Lupo, pairing onboard machine speed with human oversight shrinks the sensor-to-shooter cycle from minutes down to seconds. Yet even if a drone can analyze everything it sees in a heartbeat, a tougher question follows: how does it navigate—or alert anyone—when enemy jammers blot out every satellite in the sky?

When satellite navigation cuts out, a drone must carry its own landmarks and calculate position from within. That became an urgent reality when electronic warfare blinded navigation across entire theaters. Starting late in 2022, jamming units in Ukraine flooded front-line frequencies with enough radio noise that crews repeatedly lost their satellite signals. By the spring of 2024, GPS denial was also disrupting flight corridors above the Baltic, throughout the Eastern Mediterranean, and across the Persian Gulf.

To stay airborne without outside guidance, an aircraft relies on an internal navigation stack. An inertial measurement unit tracks motion using internal sensors, working without external signals once aligned. But calculating position purely from internal movement is like dead reckoning across a dark room: small errors compound over time into drift. Drones solve this with visual-inertial odometry, a technique that tracks how features shift frame-to-frame on camera and balances that against accelerometer data. Software fuses both streams so each sensor covers the other's weaknesses. In internal flight trials reported in 2026, one GPS-denied system held drift below 2.1 meters across a 20-minute flight.

To lock in an absolute position, the aircraft switches to terrain-relative navigation, checking live camera views against stored digital elevation models or satellite imagery. That handoff proved itself under live electronic attack during Jammertest 2025. A 10-kilogram electric vertical-takeoff aircraft detected spoofed satellite signals trying to deceive its controls. The flight computer rejected the corrupted inputs and guided the aircraft along its planned route using visual template matching.

When enemy jamming severs links with base, drones distribute targeting data using tactical mesh networking. Instead of depending on a distant command tower or satellite relay, platforms talk peer-to-peer. Transmitting across millimeter-wave radios from 24 to over 100 gigahertz, narrow directional beams pass encrypted telemetry across the formation. That self-healing web ties together air, surface, and orbital nodes without any single point of failure. With uncrewed aircraft navigating jammed skies and routing target data without turning into sovereign commanders, how does command doctrine adapt when machines execute those tactical tasks?

That distinction returns to what military doctrine actually allows. As established earlier, drones are not turning into battlefield commanders, and doctrine never lets them issue orders. Instead, they function as that rapid matching engine. As developers at Corvus Intelligence put it, edge triage is meant to direct human focus rather than supplant human judgment. Onboard models scan sensor feeds to surface recommendations, strictly bound by human-defined constraints and rules of engagement.

What shifts is how troops direct those assets. During an Army gathering in October 2025, Major General Clair Gill reportedly made clear that operators no longer need their "hands on the sticks" at every turn. Brigadier General Travis McIntosh reportedly described the true shift as directing drones by command rather than by joystick. That hinges on commander's intent—the principle where humans set the objective and boundaries, leaving tactical flight execution to software.

That concept went to sea in September 2026 off California's Point Mugu. The Navy reported testing software called CAMP on an uncrewed loyal-wingman aircraft, turning operational plans into machine instructions. Rather than flooding channels with raw video, it routes bite-sized tasks—similar to those 50-kilobyte digital postcards sent through tactical networks—letting code handle flight adjustments while humans direct intent.

Relying on intent also transforms front-line staffing. McIntosh noted that springing a drone ambush currently ties down four soldiers to fly, watch security, carry gear, and tune antennas. Lupo reportedly argued that ground forces must move past dedicated individual operators, proposing instead that a single supervisor manage fleets of up to fifty autonomous vehicles.

To protect human authority, doctrine leans on human-on-the-loop oversight—a supervisory setup where a person monitors the system and retains explicit veto power, rather than flying hands-on. Supporters point to July 2026, when DARPA, the Pentagon's research arm, announced an in-air test of an AI-piloted F-16 at Florida's Eglin Air Force Base. The pilot inside held a physical switch to seize manual control instantly. Yet that backstop begins to buckle once machines propose thousands of tactical options in seconds.

On paper, doctrine draws a hard line against pure automation. Lt. Col. Shawn Finney, who commands the Air Force's 805th Combat Training Squadron, framed it this way: “As far as human-machine teaming goes, we always will have the human as the final decision-maker.” The Joint Air Power Competence Centre defines this supervisory role as software proposing choices while a human retains authority to reject a step. Under that framework, Major Lupo argues the soldier remains the ethical anchor, required to verify positive identification—confirming a target is genuinely hostile—before weapons release.

Yet that neat separation breaks down against the clock. A cockpit toggle switch feels safe, but it collides with that question of who holds the trigger when the machine moves at light speed. In the Air Force's DASH 2 experiment, AI tools generated recommendations in under 10 seconds, producing over 6,000 solutions in an hour. Finney warned that choices arrive faster and faster, causing cognitive saturation, where an overwhelming volume of rapid algorithmic suggestions forces human supervisors into passive compliance.

Speed is only the first hurdle; the models themselves carry flaws. DARPA program manager Brigadier General James Valpiani acknowledged that hard questions remain concerning combat AI trustworthiness. Edge models face algorithmic drift, where systems trained on clean lab data degrade when exposed to real-world dust, weather, or camouflage. The Royal United Services Institute cautioned that vision algorithms can be deceived by subtle image modifications no human operator would miss, while compressing models onto compact hardware reduces accuracy and increases sensitivity to sensor noise.

When an exhausted supervisor simply rubber-stamps algorithmic choices, genuine human control disappears. Joint Air Power Competence Centre analysts caution that real-time deep learning can propose unpredictable solutions at odds with military ethics, especially since combat operations reportedly require constant care to avoid fratricide and civilian casualties. When uncritical approvals leave commanders legally accountable under the law of armed conflict for choices they had no time to evaluate, the line between supervising a tool and delegating command completely dissolves.

Under the laws of war, legal accountability rests on human shoulders. Tomorrow's drone isn't an all-powerful digital general handing down orders. It's an agile, decentralized tactical relay, processing sensor feeds out at the edge and passing targeting alerts across self-healing meshes. As Navy program manager Captain Todd Keith put it, the goal is to "reduce the burden on our operators, improve interoperability and help bring collaborative autonomy closer to operational use."

Naval doctrine is already locking those boundaries into place. On September 24, 2026, the Navy launched the Robotics and Autonomous Systems Warfighting Development Center under Rear Admiral Melvin Smith. His command is drafting tactics, techniques, and procedures for uncrewed systems, using a phased roadmap to resolve command hierarchies, mission relationships, and resource needs before autonomous aircraft join the fleet.

Those lines exist because code can't carry moral weight. Looking back at recent trials, Air Force Colonel John Ohlund noted that the military has only "grasped the edge for generating courses of action, which is just one small portion of the larger apparatus." Analysts at the Joint Air Power Competence Centre argue that command decisions are inherently sensitive and remain human, so automated suggestions can never clear human commanders of legal liability.

That boundary circles back to Lupo's ethical anchor: algorithms can propose targets, but human judgment must govern lethal swarms. Defense analysts caution that genuine human control cannot be set aside when lives are at risk. Military doctrine has pulled human hands off the flight controls. But keeping them firmly on the moral brake is the only thing separating high-speed survival from catastrophic automated error.

To keep tabs on this technology, look into human-machine teaming. And follow security; we have more episodes there so you stay ahead as defense systems evolve. Thanks for listening. Until next time, this has been Angle.