Tower 22 inquiry, base incursions show counter-drone failures fall between sensors and shooters

At about 5:30 a.m. on January 28, 2024, a one-way attack drone struck Tower 22, a U.S. outpost in northeastern Jordan, and killed three American service members. The base had counter-drone equipment. According to an Army investigation that Business Insider obtained under the Freedom of Information Act and reported in April 2025, the night shift in the base defense operations center saw two radar tracks to the south and set them aside as too distant, too slow, or possibly birds or trash. The crew was watching a ScanEagle reconnaissance drone come in to land and, in the investigation’s words, “did not interrogate or assess the unknown air tracks” with the camera available to them.

The first public explanation, offered by anonymous officials the day after the attack, was that the hostile aircraft may have been mistaken for the returning American one. The investigation as reported describes something less tidy and more instructive. Sensors produced tracks, nobody turned those tracks into an identification, and so no decision was ever made. Investigators also assessed that the crew “was not properly manned with the appropriate rank and experience level.”

Nothing in that account is a hardware failure, yet counter-drone capability is still bought, tested and written about one box at a time: a radar with a stated range, a jammer with a stated band, an interceptor with a stated price. The useful unit of analysis is the whole chain from detection through tracking, identification, decision and defeat, together with the people and legal authorities that connect each step to the next. A program that reports only component performance is reporting on the parts of the problem that were already the easiest.

A track on a screen is not an identification

The Government Accountability Office made the basic technical point in a March 2022 technology assessment: radar and radio-frequency sensors are the most common detection methods, and “electromagnetic interference (e.g., power lines and LEDs) and small airborne objects (e.g., birds) can decrease detection capabilities or generate false detections.” Tower 22 shows the human consequence. A sensor that regularly shows clutter teaches its operators to discount what it shows, and a crew that has learned to discount tracks will eventually discount a real one.

The same gap sits between identification and action. A classifier that labels a track hostile with high confidence has produced an input. Someone with the authority, the training and the time still has to act on it, through an interface that either helps or gets in the way. Two officers of the 10th Mountain Division’s 2nd Brigade Combat Team, which defended bases in Iraq and Syria against more than 170 attacks between August 2023 and April 2024, wrote afterward that early software required radar operators to perform “up to fourteen clicks of a mouse to interrogate suspected tracks and deploy countermeasures.” They also found interceptors “self-detonating or refusing to launch altogether based on default engagement settings.” Neither problem appears on a specification sheet, and both sit squarely between a correct detection and a defeated drone.

Cost per shot is a component metric too

The Red Sea campaign is usually told as a story about expensive missiles. In January 2025 the commander of Naval Surface Forces, Vice Adm. Brendan McLane, said that over 15 months Navy ships had fired 120 SM-2s, 80 SM-6s, 20 Evolved Sea Sparrow and SM-3 missiles and 160 five-inch gun rounds against nearly 400 Houthi drones and missiles. An SM-2 costs roughly $2.1 million. The response has been a search for cheaper effectors, and one has been demonstrated in combat. Gen. Michael Kurilla, the head of U.S. Central Command at the time, said laser-guided APKWS rockets fired from F-16s and F-15s accounted for about 40 percent of Houthi drone kills during Operation Rough Rider, in remarks reported in June 2025: “That’s a $25,000 munition going against a roughly $50,000 or $100,000 drone.”

That is real progress, and it is easy to misread. The rocket is cheap because the fighter, its targeting pod, the tanker and the air picture that put it behind the drone were already paid for. Erik Schuh, an Air Force operations research analyst, argued in November 2025 that the true cost of an intercept includes “the destroyer that fires them, the crew that mans the system, the fuel and maintenance to keep it at sea.” Wes Rumbaugh of CSIS had made the complementary case in February 2024 that the value of what is defended belongs in the calculation. Cost per shot is one more single-component number. A low-cost interceptor paired with a sensor that cannot cue it, or a crew that cannot identify the target in time, has an attractive price and no effect.

At home, the missing link is authority

Inside the United States the weakest link has been legal. Under 10 U.S.C. 130i, as the Congressional Research Service explains, the department may detect, track and mitigate drones only at “covered” facilities tied to specified missions, in coordination with the Federal Aviation Administration and with spectrum-emitting equipment subject to Federal Communications Commission evaluation. When drones flew over Langley Air Force Base on 17 days in December 2023, the Pentagon later confirmed the incursions and said the aircraft “didn’t appear to exhibit any hostile intent.” Their origin was not determined. Gen. Gregory Guillot of U.S. Northern Command told senators in February 2025 that there had been 350 detections over 100 installations in the previous year and that only about half of installations were covered.

The department’s inspector general found the paperwork as much at fault as the statute. Its January 2026 advisory, as reported by Defense News, counted more than 20 separate policies, noted that Luke Air Force Base, where most F-35 pilots train, “is not designated as a covered facility or asset,” and said “a large percentage of installations” lacked operational approval to use what they had. A policy signed on December 8, 2025 let service leaders designate covered facilities and treats unauthorized drone surveillance of an installation as a threat. A memo signed on September 28, 2026 goes after spectrum approvals, authorities to operate, weapons safety reviews and legal reviews. “The Department will take down barriers so our warfighters can take down drones,” Defense Secretary Pete Hegseth said.

These are chain-level fixes, and the best available outcome figure shows how far there is to go. Guillot testified in March 2026 that a deployed fly-away kit had detected and defeated small drones over a strategic installation in the opening hours of Operation Epic Fury. He added: “Whereas a year ago, almost every one that was detected was not defeated. Now about a quarter of the ones that we detect we’re able to defeat.” Three in four detected drones still complete their flight.

Test the chain where it will be used

The organization responsible has changed. The Joint Counter-small UAS Office, created in 2020, evaluated 40 interim systems and endorsed eight, but CRS records congressional concern that the services were not buying what it recommended. In August 2025 Hegseth replaced it with Joint Interagency Task Force 401, whose director has procurement authority of up to $50 million per effort and which absorbed the resources of Replicator 2, the installation-defense initiative announced in September 2024. The first Replicator 2 purchase, announced in January 2026, was two sets of Fortem DroneHunter F700 net-capture interceptors.

The task force’s more consequential work is on evaluation. Its test standards were first applied in Lithuania in May 2026, where more than 20 systems were assessed with instrumented data and operator feedback, and in September it set up permanent test lanes at Yuma Proving Ground. Common criteria are overdue. The 10th Mountain officers reported that live-fire experiments “gave a false sense of confidence” because the friendly target drones “were easier to shoot down than the enemy drones.”

A common standard can still measure the wrong thing. A result from the Arizona desert says little about a shipyard surrounded by civilian emitters, and a trial against drones flying in from a distance says little about the geometry Ukraine used on June 1, 2025, when 117 first-person-view drones were launched from trucks parked near five Russian air bases. Janes verified 12 aircraft destroyed. A defense built around long-range warning has almost no time to work against a launch from just outside the fence.

Three changes follow. Test reports, whether from a vendor or from Yuma, should state the conditions and the chain-level outcome: time from first track to a declared identification, false tracks per shift, the crew size and rank assumed, and the authority the operator was presumed to hold. Acceptance testing should be repeated at the site and with the crew that will use the system. Finally, the annual public report on use of 130i authority that the fiscal 2026 defense authorization act requires should carry Guillot’s ratio of drones detected to drones defeated, installation by installation where classification allows. That number captures the sensor, the operator, the lawyer and the interceptor at once, which is why it is the one worth publishing.