In late January 2026, a Russian BM-35 strike drone flew over the government quarter of Kyiv. It was being steered through a Starlink satellite terminal, a link Ukrainian jammers could not easily touch. Ukraine’s defense minister, Mykhailo Fedorov, took the evidence to SpaceX.
According to The Atlantic, the company built a whitelist of authorized Ukrainian terminals in about a day. By early February, unregistered terminals along the front had gone dark, and the Institute for the Study of War reported that the restrictions were disrupting Russian drone operations.
A software change sharply degraded the operational advantage Russia had built around Starlink in days. Strictly speaking, this was access control on a satellite network, outside classic electronic warfare. That is the point: the contest around spectrum-dependent systems now extends into software, network access, and autonomy.
The classic jamming contest moves almost as fast. On Ukraine’s front lines, electronic warfare defeats a capability, a countermeasure emerges, better electronic warfare defeats it, and the cycle repeats within weeks, according to an analysis published by West Point’s Modern War Institute.
In the electromagnetic spectrum, military advantage has become perishable. It can last weeks, sometimes days.
A dependency that grows with every sensor
The spectrum has always mattered in war. What has changed is how much of a modern force runs through it. Drones, radars, tactical networks, satellite links, GPS-guided munitions, and command-and-control systems all depend on the electromagnetic spectrum for some combination of sensing, communication, and navigation.
If the battlefield operating layer connects sensors, data, decisions, and effectors, as argued in this column in July, the spectrum is the medium those connections travel through. The more connected and intelligent a force becomes, the more electromagnetic dependencies it creates.
The Middle East offers a measure of that exposure. In the months after the war with Iran that began on February 28, over one million GPS jamming incidents were recorded in the Gulf, 98% of all incidents worldwide, according to maritime intelligence firm Windward, as reported by Sky News.
Between February 28 and March 3 alone, Lloyd’s List Intelligence logged 1,735 interference events affecting 655 vessels. Ships, aircraft, and delivery drivers across the region found their positions displayed at sea, on land, or in the wrong city.
Adaptation moves between layers
Ukraine shows what adaptation looks like in practice. Each countermeasure pushes the other side to change a different layer of the system.
The first layer is the signal itself: frequencies, waveforms, and firmware. Here, the Modern War Institute analysis describes three distinct tempos. The jamming-and-countermeasure cycle repeats within weeks. Field modifications developed by brigade innovation teams reach the next production run within days or weeks.
Manufacturers release formal software and hardware updates several times per year. The Pentagon’s major defense acquisition programs take more than 12 years on average to deliver a capability, according to the Government Accountability Office.
The second layer is the link. Fiber-optic drones, guided through a thin cable that unspools behind them, emit no radio signal for a jammer to find. By April 2026, they accounted for 32% of Ukraine’s strike drones, according to figures cited by the Institute for the Study of War. The cable limits range and can snag on terrain, so the protection comes at a price.
The third layer is navigation. Developers are fielding AI-based visual navigation, such as KrattWorks’ Ghost Dragon, designed to keep drones operating in GPS-denied and jammed conditions.
When one layer becomes vulnerable, adaptation moves elsewhere: from frequency to firmware, from radio links to fiber, and from GPS to onboard autonomy.
From programmed systems to learning systems
Traditional electronic warfare relies heavily on threat libraries. Analysts catalog the parameters of known radars and radios, engineers develop a countermeasure for each, and the result is loaded into the system before it deploys. When an adversary fields a new waveform, the cycle goes back to the lab. That process can take months, an interval the war in Ukraine has made untenable.
AI compresses the chain: detect, classify, identify, decide, reconfigure. AI research is pushing toward systems that can characterize previously unseen signals and generate countermeasures while an encounter is still unfolding. One early effort by the US Defense Advanced Research Projects Agency (DARPA), the Behavioral Learning for Adaptive Electronic Warfare (BLADE) program, demonstrated the principle.
In tests, BLADE detected, characterized, and countered new wireless communication threats in minutes rather than months, according to program contractor Lockheed Martin. The US Army has run competitions and field experiments on AI-based classification of unknown signals – aimed at helping soldiers make sense of a crowded spectrum.
These remain research programs and experiments, and fielding them at scale is a harder problem. The strategic implication is already visible. AI is compressing the time between a new signal and a new response. As that interval shrinks from months toward minutes, adaptation itself becomes a capability exercised during the fight.
AI on both sides of the contest
AI does more than sharpen jammers. It helps militaries understand the spectrum, classify signals, and choose countermeasures. It also reduces how much a system needs the spectrum in the first place.
A 2025 report by the Center for Strategic and International Studies estimated that Ukrainian drones with AI-enabled autonomous navigation raised target engagement success rates from 10%-20% to around 70%-80%. A human still selects the target. The autonomy handles the final approach, so the strike no longer depends on a stable communications link.
AI helps militaries attack an adversary’s electromagnetic dependence, and it helps their own systems keep operating when links are jammed. Both sides are using the same technology to make the other’s spectrum weapons less relevant.
The radio-frequency environment is adversarial by design. Opponents deceive deliberately, new signals may look nothing like the training data, connectivity can vanish, and a misclassified emitter can lead to a wrong and dangerous decision.
Adoption is also uneven: of nearly two million drones Ukraine contracted in 2024, CSIS counted about 10,000 with confirmed AI guidance. The real task is building AI that performs in novel, degraded, and deceptive conditions, with clear human accountability for what it does.
Israel’s spectrum front
Israel has already tested this in combat. During the 12-day war with Iran in June 2025, the IDF’s 5114th Spectrum Warfare Battalion thwarted dozens of drone infiltration attempts using electronic warfare, the military announced at the time. Military officials later told The Times of Israel that the battalion took out about 25% of the roughly 1,100 drones Iran launched at Israel.
The spectrum array had been built as a defensive capability, and during the war, the IDF found it could be used offensively as well. In December 2025, the IDF completed a reorganization of its C4I and Cyber Defense Directorate that created a Spectrum and Communications Division and an Information and Artificial Intelligence Division.
Some aerial threats can be defeated without firing a missile. Others cannot be defeated electronically at all. Air defense is becoming layered across interceptors, lasers, electronic warfare, and deception, and each layer needs its own adaptation cycle.
For Israel, strengthening that cycle requires two things: First, spectrum data, from signal libraries to interference logs, should be managed as a national asset and shared quickly with industry under appropriate security and classification controls, as part of a broader effort to turn wartime data into defensive AI.
Second, procurement should fund continuous model and firmware updates, and suppliers should be measured by how fast they can field one. That speed is the practical meaning of execution at scale in the spectrum.
A contest between learning systems
Electromagnetic superiority now has an expiry date, and it has to be regenerated continuously. In the spectrum, the learning loop is increasingly moving from between engagements into the engagement itself.
Militaries count sorties, interceptors, and stockpiles. In the spectrum, the number that may matter most is the time between an adversary’s change and an effective response.
For decades, electronic warfare was largely a competition between systems: a jammer against a radar, a signal against a counter-signal. AI is beginning to turn it into a competition between learning systems.
This analysis is based solely on open-source information and publicly available reporting.