For most of the 20th century, electronic warfare was a contest of libraries. The side that could pre-load more radar signatures, more communication waveform templates, and more jamming techniques into its systems before a conflict began held the advantage. EW operators were essentially curators of electromagnetic dictionaries — the better your pre-mission intelligence about enemy systems, the more effectively you could jam, spoof, or exploit them.

That paradigm is dead. The electromagnetic spectrum is now contested by software-defined radios that can change their emission characteristics in milliseconds, frequency-hopping waveforms that make traditional jamming ineffective, and cognitive radar systems that learn to defeat interference in real time. Against adversaries that can change their electromagnetic signature faster than human analysts can characterize it, static EW libraries are useless. The battlefield demands systems that can learn to jam in real time — and that is exactly what AI-driven electronic warfare is designed to deliver.

<50ms Target reaction time of next-generation cognitive EW systems — faster than human neural response to perceived threat

What Is Cognitive Electronic Warfare?

Cognitive Electronic Warfare (CEW) refers to EW systems that use machine learning and AI to perceive their electromagnetic environment, reason about threat characteristics, and adapt their jamming or deception techniques in real time — without pre-programmed responses and without human intervention in the loop. The key distinction from conventional EW is generalization: a cognitive EW system can generate effective responses to signal types it has never encountered before, by reasoning from first principles about the physics of electromagnetic propagation and interference.

The architecture of a cognitive EW system typically includes three layers. The first is a wideband sensing layer — a software-defined radio front end capable of monitoring multiple gigahertz of spectrum simultaneously, capturing raw signal data. The second is an inference layer — a machine learning model, often a combination of convolutional neural networks for signal classification and reinforcement learning for jamming strategy selection, that interprets the captured signals and identifies threats. The third is the response layer — a software-defined transmitter that executes the chosen jamming, spoofing, or deception strategy.

What makes this architecture transformative is the feedback loop. After executing a jamming technique, the system can observe whether the target radar or communication system modified its behavior — whether it changed frequency, adjusted power, switched waveforms — and update its strategy accordingly. This is machine learning applied to electronic warfare: the system improves with each engagement, building an increasingly accurate model of adversary EW systems and how to defeat them.

DARPA's Foundational Programs: BLADE and ARC

The conceptual and technical groundwork for AI-driven EW was laid at DARPA through two critical programs: the Behavioral Learning for Adaptive Electronic Warfare (BLADE) program and the Adaptive Radar Countermeasures (ARC) program. Together, these represent the intellectual lineage of every current U.S. cognitive EW system.

BLADE: Learning to Jam Communications

BLADE, which entered development in the early 2010s with Lockheed Martin as the prime contractor, tackled one of the hardest problems in EW: how to effectively jam new wireless communications threats that have never been encountered before. The traditional approach — loading pre-built jamming techniques into a library and triggering them on recognition — fails completely against novel waveforms. BLADE's approach was radically different: rather than matching a threat to a pre-built response, the system would autonomously detect the new threat, characterize its waveform properties, generate candidate jamming techniques, test them, assess effectiveness, and refine — all in real time.

The system was designed to operate as a single node or as a network of distributed BLADE nodes, with performance improving as nodes were added — a distributed learning architecture that presaged later concepts like federated learning for military applications. Field tests demonstrated BLADE achieving effective jamming of novel waveforms within seconds of first encounter, compared to days or weeks in the traditional intelligence-analysis-programming pipeline.

ARC: Defeating Cognitive Radars

If BLADE addressed communications jamming, ARC targeted an equally critical domain: radar countermeasures against digitally programmable, agile-waveform radar systems. Modern military radars don't transmit on a fixed frequency with a fixed pulse shape — they use frequency hopping, pulse compression, and waveform agility specifically to defeat traditional jamming. ARC, described by DARPA as "jamming carried out against radars with unknown behaviors and agile waveform characteristics," used AI to characterize unknown radar waveforms and generate tailored countermeasures.

The ARC program's key insight was that even an unfamiliar radar system must obey the laws of physics — it has constraints on its waveform design, its receiver characteristics, and its signal processing chain that can be inferred and exploited. An AI system trained on the physics of radar design can generate novel jamming waveforms that exploit these constraints even for radars it has never seen before. This is cognitive warfare in the truest sense: fighting at the level of physics, not just pattern matching.

⚠ PROGRAM STATUS: DARPA EW PORTFOLIO 2026

DARPA's current EW portfolio has expanded well beyond BLADE and ARC. The Adaptive Electronic Warfare (AEW) program is developing cognitive EA (Electronic Attack) systems for contested environments. The Radio Frequency Machine Learning Systems (RFMLS) program — transitioned from DARPA to Air Force Research Laboratory — is focused on automated signal classification and anomaly detection for EW applications. The Computational Simulation of Online Adaptive Learning (CSOA) program is building training environments for AI EW systems that can simulate millions of engagement scenarios. Combined budget across these programs exceeds $800 million as of FY2025.

L3Harris and the Commercial Integration of Cognitive EW

DARPA's research programs are designed to transition into fielded capabilities, and L3Harris Technologies has emerged as one of the primary commercial integrators of cognitive EW subsystems. Their approach — described in published technical literature as building compact, self-contained "cognitive EW modules" — reflects a philosophy of modular intelligence: rather than designing complete EW systems with AI built in from scratch, they build AI-capable signal detection and response modules that can be integrated into existing platforms.

L3Harris cognitive EW modules are designed to respond in real time to unrecorded and unknown signal waveforms — the specific capability that BLADE and ARC were designed to develop. The modules incorporate digital wideband receivers, onboard AI inference engines, and software-defined transmitters in a package compact enough to integrate into aircraft, ships, ground vehicles, and even large UAS platforms. By building modular cognitive EW capability, L3Harris enables platform-agnostic deployment of AI jamming capabilities across the force structure.

The commercial significance of this approach extends beyond military procurement. Cognitive EW modules developed for military applications are directly analogous to cognitive radio technology with broad applications in telecommunications, satellite communications, and contested spectrum environments — creating a commercial technology market that drives investment and innovation beyond the defense sector alone.

Spectrum Warfare: The New High Ground

The electromagnetic spectrum has always been a military domain, but AI is accelerating the pace of spectrum competition to a point where human cognition can no longer keep up. Modern military operations depend on the spectrum for everything: GPS navigation, drone control links, satellite communications, radar, targeting, and command-and-control networks. Whoever can dominate — or deny — the spectrum in a given operational area holds a decisive advantage across every other military domain.

The 2024 U.S. Joint Electromagnetic Spectrum Operations (JEMSO) doctrine update explicitly recognized AI as a critical enabler of spectrum dominance, designating "Electromagnetic Spectrum Operations" as a standalone warfighting function equal in importance to maneuver, fires, and sustainment. This doctrinal shift reflects the operational reality that electronic warfare is no longer a support function — it is a primary form of combat.

$4.7B U.S. Department of Defense EW budget FY2025 — largest in history, 23% increase from FY2024

Multi-Domain EW Integration

AI-driven spectrum warfare is not confined to dedicated EW platforms. Modern combat aircraft are increasingly equipped with integrated electronic attack capabilities — the F-35's AN/ASQ-239 Barracuda electronic warfare system, for example, provides full-spectrum sensing and active electronic attack in a package integrated into the aircraft's primary mission systems. The logical next step, already in development, is connecting these distributed EW nodes into a mesh network where AI coordinates spectrum operations across multiple platforms simultaneously.

The concept is called Collaborative Electronic Attack (CEA): a group of platforms — aircraft, ground vehicles, ships, or UAS — each equipped with sensors and transmitters, coordinated by AI to generate coherent jamming effects across wide geographic areas. Each platform acts as a node in a distributed EW network, with the AI managing beam forming, frequency allocation, power levels, and timing to maximize jamming effectiveness against specific threat radars or communication systems. This approach is architecturally similar to how phased-array radars work — multiple transmitters coordinated to create precisely shaped electromagnetic effects — but applied to electronic attack rather than sensing.

Signals Intelligence Automation: SIGINT at Machine Speed

Electronic warfare's offensive cousin is Signals Intelligence (SIGINT) — the collection, processing, and exploitation of adversary electromagnetic emissions for intelligence purposes. Traditionally, SIGINT required enormous human analytical effort: intercept operators listening to communications, signals analysts characterizing radar waveforms, and cryptanalysts working to break communications security. AI is automating all of these functions at a scale and speed that is transforming intelligence collection.

The NSA's GHOSTMACHINE program (referenced in Edward Snowden documents) was an early example of AI-assisted SIGINT processing, designed to automate the classification and routing of intercepted signals. By 2025, AI-driven SIGINT processing has advanced dramatically. Modern systems can automatically classify intercepted signals by type (radar, communications, navigation), identify specific emitter platforms by their distinctive "fingerprint" (Electronic Intelligence, or ELINT), transcribe and translate voice communications, extract entities and relationships from text communications, and flag anomalies that warrant human analyst attention.

The throughput advantage is staggering. A human SIGINT analyst can process perhaps a dozen intercepts per hour with full analytical attention. An AI system running on modern GPU hardware can process millions of intercepts per hour, flagging the handful that require human review. This asymmetry means that AI-enhanced SIGINT collection can maintain comprehensive electromagnetic awareness of an adversary's entire communication and radar network with analytical staffing that would previously have only been sufficient to monitor a single frequency band.

Counter-Drone EW: The Ukraine Crucible

The most immediate real-world testing ground for AI-driven EW has been the drone war in Ukraine. By 2024–2025, both Ukrainian and Russian forces were deploying FPV drones by the hundreds of thousands, and both sides were investing heavily in electronic warfare systems to counter them. The resulting technological competition has driven innovation in counter-drone EW at a pace never seen in peacetime development.

Traditional counter-drone jamming works by flooding the frequency bands used by drone control links with noise, severing the connection between operator and drone. The drone then either hovers in place (awaiting reconnection), returns to its last known operator location, or crashes — depending on its programmed failsafe behavior. Against early-generation FPV drones operating on fixed frequencies, this approach was effective. Against modern FPV drones with frequency-hopping control links, spread-spectrum communications, or fiber-optic tethers, traditional jamming is largely ineffective.

The response has been cognitive counter-drone EW: systems that can identify the specific waveform characteristics of a given drone's control link, determine its frequency-hopping pattern, and execute targeted jamming that follows the frequency-hopping sequence. This requires real-time signal analysis and response capabilities that are only feasible with AI — the control link may hop hundreds of times per second, faster than any human operator can track.

Ukrainian forces have deployed AI-powered counter-drone systems, including the domestically developed Kvertus Electronic Warfare complex, which uses machine learning to characterize drone control signals and adapt jamming responses. By late 2025, Ukrainian manufacturers had also developed fiber-optic controlled FPV drones specifically to defeat RF-based counter-drone EW — a technical adaptation that rendered entire generations of jamming equipment obsolete and forced another round of innovation from Russian EW developers.

⚠ TECHNICAL NOTE: THE FPV/EW ADAPTATION CYCLE

The Ukraine drone war has demonstrated a technology adaptation cycle that has compressed from years to weeks. When Russian EW systems effectively jammed standard 5.8GHz FPV control links, Ukrainian drone developers switched to encrypted control protocols. When Russia developed wideband jamming to cover multiple frequency bands, Ukraine introduced fiber-optic tethered FPVs with 20km range. When Russia deployed AI-assisted signal classification to identify fiber-optic drone acoustic signatures, Ukraine fielded thermal-masked ground stations. Each cycle lasts approximately 6–8 weeks before both sides have adapted. AI has compressed what was once a multi-year R&D cycle into a combat-driven sprint — with implications for how all future technology competition in contested electromagnetic environments will unfold.

Generative AI in EW: The AOC 2025 Turning Point

The 62nd Annual AOC International Symposium & Convention — the premier gathering of electronic warfare professionals — chose generative AI and cognitive EW as its central technical focus. This choice reflected a significant shift in community thinking: generative AI, which had previously been viewed primarily as relevant to natural language processing and image generation, is now understood to have direct applications in electronic warfare.

The specific application is waveform generation. Generative AI models — particularly Generative Adversarial Networks (GANs) and transformer-based architectures — can be trained to generate novel electronic attack waveforms that are specifically optimized to defeat target radar and communications systems. Rather than selecting from a library of pre-built jamming techniques, a generative AI EW system can synthesize a novel waveform tailored to the specific characteristics of the threat it is facing, in real time.

This capability has been demonstrated in DARPA-funded research using reinforcement learning: an AI agent controlling a software-defined radio played adversarial games against another AI controlling a target radar system, with the EW agent learning increasingly sophisticated jamming strategies through millions of simulated engagements. The resulting strategies were often non-intuitive to human engineers — novel approaches that exploited subtle vulnerabilities in the target system's signal processing chain that no human analyst had identified.

The Dual Problem: Defending AI-Driven EW Systems

As cognitive EW systems become widespread, they also become targets for adversarial attack. An AI-driven EW system can potentially be deceived by adversarial signals specifically crafted to cause misclassification — feeding the system false data about threat characteristics to induce incorrect jamming responses. This adversarial AI attack on EW systems is an emerging research area with significant operational implications.

Russian military researchers have published (in open-source journals) work on "adversarial waveform generation" — techniques for crafting electromagnetic emissions that deliberately confuse AI-based signal classifiers. The practical application would be a radar or communication system that, by subtly modulating its emissions in a specific pattern, causes an adversary's AI EW system to misidentify it as a non-threat or to apply an ineffective jamming response. This is the EW application of the same adversarial examples that fool image classifiers into misidentifying stop signs as speed limit signs.

The implication is a layered AI arms race within electronic warfare: cognitive EW systems that learn to jam must also be hardened against adversarial inputs, which requires AI-based anomaly detection in the sensing layer, robust training against adversarial examples, and human-in-the-loop verification for high-stakes jamming decisions. The complexity of this adversarial dynamic — AI systems attacking AI systems attacking AI systems — is increasing faster than current governance frameworks can track.

China and Russia: Peer Competitor Cognitive EW Development

The United States is not alone in developing AI-driven EW capabilities. China's PLA Strategic Support Force, responsible for EW, has published extensively on cognitive EW research, and multiple Chinese universities have active programs in AI-based signal processing and electronic countermeasures. The specific technical approaches mirror DARPA's work: reinforcement learning for jamming strategy optimization, deep learning for signal classification, and GAN-based waveform generation.

Russia's cognitive EW development has followed a different path — driven more by operational necessity in Ukraine than by pre-war research investment. The Krasukha-4 broadband multifunctional jamming station and its successors have been updated with AI-assisted signal processing modules. Russian ELINT assets have reportedly incorporated machine learning to accelerate the characterization of Ukrainian and NATO communication systems. The pace of Russian EW innovation, surprisingly, has been sustained by the Ukrainian conflict despite significant resource constraints elsewhere in the Russian military system.

// KEY INTELLIGENCE POINTS
  • Cognitive EW systems respond to unknown threats in <50ms — DARPA BLADE and ARC programs established foundational capabilities, now in operational transition
  • L3Harris modular cognitive EW architecture enables platform-agnostic AI jamming deployment across all force elements
  • DoD EW budget reached $4.7B in FY2025 — largest in history, with AI/ML integration as primary growth driver
  • Ukraine drone war has compressed EW innovation cycles from years to weeks — adversarial adaptation is now real-time
  • Generative AI is enabling novel attack waveform synthesis that no human analyst could design — AOC 2025 recognized this as a doctrinal shift
  • Adversarial attacks against AI EW systems represent an emerging asymmetric threat — cognitive EW creates cognitive EW vulnerabilities
  • China's PLA Strategic Support Force has active parallel programs mirroring U.S. DARPA EW research portfolio

The Spectrum as a Weapons System

The electromagnetic spectrum has always been described as a military domain, but AI is making it a weapons system in its own right. A cognitive EW network can degrade an adversary's military effectiveness across every physical domain — by jamming aircraft radar, it creates opportunities for air defense penetration; by disrupting drone control links, it removes precision strike capability; by spoofing GPS signals, it degrades precision navigation; by intercepting and exploiting communications, it creates battlefield intelligence that enables targeted strikes. AI doesn't just make EW more efficient — it makes spectrum dominance a force multiplier for all other military operations simultaneously.

For military planners, the implication is that electronic warfare is no longer a supporting capability to be assigned to specialist units and consulted when needed. It is a primary warfighting function that must be integrated at every echelon, from the individual soldier's radio to the carrier strike group's electronic attack aircraft. And the AI that enables cognitive EW is increasingly available to every actor with access to modern commercial computing hardware and open-source machine learning frameworks — which means the threshold for deploying basic cognitive EW capabilities is falling to a level accessible to non-state actors and smaller militaries.

The spectrum war is not coming. It is here. The question is not whether AI will transform electronic warfare — it already has. The question is whether the force structures, doctrine, training pipelines, and acquisition systems of Western militaries are adapting to that transformation fast enough to maintain the advantage.