The first flight test of an autonomous electronic warfare ecosystem took place on July 16, 2026, on a test range in the United States. An unmanned aircraft detected, analyzed, and responded to electromagnetic threats without a single human command. The two companies flew the integrated Distributed Spectrum Collaboration and Operations (DiSCO) ecosystem with Hivemind mission-autonomy software aboard a Green Wolf launched-effects platform.
The milestone compresses what DARPA began building in 2010 under the BLADE and ARC programs: behavioural learning for adaptive EW and adaptive radar countermeasures. The live-flight demonstration moved from simulation in February 2026 to a flight test five months later.
The test shifts the electromagnetic spectrum from contested terrain where the faster jammer wins, to intelligent terrain where the faster learning algorithm wins.
The US Congress authorised $44.3 million for cognitive electromagnetic warfare research in the FY 2026 NDAA. The Pentagon's EW budget reached $4.7 billion in FY2025, with AI and ML integration as the primary growth driver.
Broader cognitive EW systems, including hardware, integration, and training, are forecast to grow from $27.5 billion to $63.3 billion over the same period (14.8% CAGR), driven by the shift from static threat libraries to machine-learning architectures.
From libraries to learning
For six decades, electronic warfare followed one playbook: detect a signal, match it against a pre-loaded threat library, fire the corresponding jamming technique. The system was only as good as its last intelligence update. When Russian and Chinese forces fielded software-defined radios capable of changing waveforms in milliseconds, using frequency-hopping patterns that shift faster than a human operator can characterize, that playbook became obsolete.
DARPA recognised the gap early. BLADE (Behavioural Learning for Adaptive Electronic Warfare, awarded to Lockheed Martin) proved that a machine-learning system could encounter a novel communications waveform, characterise its properties, generate candidate jamming techniques, test them, and refine its approach, all within seconds. ARC (Adaptive Radar Countermeasures, led by BAE Systems under a $35.5 million cumulative contract) tackled the harder problem of defeating cognitive radars that themselves adapt to interference.
The transition from DARPA research to operational capability has accelerated sharply since 2024. The Army Research Laboratory's FREEDOM programme (Fundamental Research for Electronic Warfare in Multi-Domain Operations) is funding RF sensing architectures, AI models optimised for embedded inference, and closed-loop EW techniques for autonomous operation across contested spectrum.
What cognitive EW does that conventional EW cannot
The key difference is response latency. Traditional EW systems require minutes to hours for human analysts to characterise a new signal and update libraries. Cognitive EW systems respond in under 50 milliseconds, faster than human neural response to a perceived threat.
Three concurrent deployments
The L3Harris-Shield AI flight test is not an isolated experiment. Three Army programmes illustrate how far cognitive EW has moved beyond the lab:
BEAST+. The Mastodon Beast+ system, a multi-channel EW device for detection, direction-finding, and attack, deployed with 3rd Infantry Division units during Exercise Combined Resolve 25-02. Soldiers used it to detect signals across the spectrum, identify enemy positions, and apply countermeasures through an AI-driven interface that fused data from multiple sensors.
REWSI. The Army's Rapid Electromagnetic Warfare and Signals Intelligence Commercial Solutions Offering, launched in June 2026, functions as a curated library of pre-vetted commercial cognitive EW capabilities. Commanders can pull off-the-shelf AI and ML modules into theatre at the speed adversaries change their electronic order of battle, bypassing the traditional multi-year acquisition cycle.
Spectrum Situational Awareness System. This capability gives commanders real-time visualisation of their own unit's electromagnetic signature, enabling signature management at the formation level. It treats EW not as an isolated function but as a continuous operational discipline.
Each of these programmes runs on a different acquisition timeline and serves a different echelon, but they share a common architectural assumption: static threat libraries are dead. Every new-start EW programme since 2023 has adopted cognitive architecture as the default.
The industrial base lines up
Every major Western defence prime now operates a cognitive EW programme. BAE Systems completed Phase 3 of ARC, transitioning cognitive countermeasure algorithms to fifth-generation fighter platforms. Northrop Grumman won a $200 million US Navy contract for AI-driven EW systems in 2024. Raytheon demonstrated the first AI-powered Radar Warning Receiver for fourth-generation aircraft in February 2025. Leonardo partnered with UK-based Faculty AI on the Cognitive Intelligent Sensing (CoInS) framework.
Thales published a detailed architecture white paper in April 2026 covering AI-integrated electromagnetic warfare. "The key issue with this technology will be making sure we do not become dependent on others," said Frantz Loutrel of Thales, framing cognitive EW as a sovereignty question.
Lockheed Martin opened a dedicated cognitive EW R&D facility in Maryland and is actively hiring AI and ML engineers for signal-processing applications across radar, communications, and EW.
The adversarial dimension
Cognitive EW creates a new layer of vulnerability. If an AI-driven jamming system can be deceived by adversarial signals crafted to cause misclassification, feeding it false data about threat characteristics to induce incorrect responses, then the EW fight becomes a machine-speed contest of adversarial learning.
The practical application would be a radar or communication system that, by subtly modulating its emissions in a specific pattern, causes an adversary's cognitive EW system to misidentify it as a non-threat or 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, strong training against adversarial examples, and human-in-the-loop verification for high-stakes jamming decisions.
The Ukraine conflict has compressed EW innovation cycles from years to weeks. Adversarial adaptation between Ukrainian and Russian EW systems is now real-time. The L3Harris-Shield AI flight test demonstrates that the United States is fielding autonomous, adaptive, learning capability rather than relying on libraries that take months to update.
Generative AI and the EW arms race
What happens to the electromagnetic battlefield a year from now?
Probability: 70% — The technical bottleneck is no longer hardware. Software-defined radios with sufficient processing headroom for onboard ML inference are commercially available. The bottleneck is validation: verifying autonomous EW behaviour before operational deployment. The L3Harris-Shield AI flight test demonstrates that validation path.
✅ Arguments for
Confirmation criteria: A contract award for cognitive EW retrofit of a non-frontline platform (transport aircraft, ground-based air defence) within 12 months.
❌ Arguments against
Disconfirmation criteria: A documented adversarial AI evasion of a fielded cognitive EW system causing operational failure.
Development scenarios
🟢 Optimistic scenario (30%)
Implications: The DoD EW budget allocation shifts from sustaining legacy libraries to procuring cognitive modules, creating a multi-billion dollar procurement cycle for the startup base.
🟡 Base-case scenario (50%)
Implications: Growth in the cognitive EW market tracks current projections ($1.44 billion by 2030), with a narrow supplier base dominated by primes.
🔴 Pessimistic scenario (20%)
Implications: Static-library systems remain the operational standard through the late 2020s. VC interest in defence AI EW narrows to startups offering hybrid human-AI architectures rather than fully autonomous systems.
Growing: VC investment in defence AI — on pace to surpass $18 billion in 2026. The startup pipeline for cognitive EW-specific capabilities (Constelli, Epirus) is accelerating.
Falling: The traditional threat-library model. No major Western defence programme has initiated a new static-library EW development since 2023. Cognitive EW is the default architecture for all new starts.
Emerging: University spinouts in cognitive EW signal processing. The SBIR and STTR pipeline now includes active topics for AI-based spectrum sensing, Rydberg atomic sensor integration, and autonomous EW behaviour validation.
Watch: The companies plan additional flight tests with expanded mission sets through open architecture for EW at the tactical edge. The transition from flight test to production contract will signal whether this capability is heading for program-of-record status.
