Air Forces Analysis Artificial Intelligence Defense Industry Technology

How AI Is Changing Electronic Warfare in Europe: Cognitive EW

AI in electronic warfare means using machine learning to recognise radar and radio signals that are not in a pre-loaded threat library, to build a live picture of the spectrum and to propose responses faster than an operator can. In Europe the first programme to put that on a front-line fighter is Germany’s Eurofighter EK, where Saab’s Arexis suite will carry Helsing’s Cirra software under €549 million of orders placed in November 2025, with 15 retrofitted jets due to be operational in 2030. Thales, Leonardo, Hensoldt, Elettronica, Indra, Rohde & Schwarz, BAE Systems and ASELSAN are all building “cognitive EW” features, but most remain demonstrators rather than fielded systems.

Key facts (as of September 2026)

Eurofighter EK: Airbus ordered Arexis from Saab in two tranches, €291 million and €258 million; the second includes the Saab/Helsing AI platform. Deliveries run to 2028, operational use from 2030.
Helsing’s Cirra deep-learning software is being integrated into Arexis over three years after an 18-month concept phase and evaluation flights on a company-owned test aircraft.
Saab added a second AI partner in September 2026: Finnish firm NestAI, to let parts of Arexis keep learning after delivery.
EU money: the European Defence Fund is financing REACT II (€40 million), SCEPTER (€35 million) and AI-WASP (€45 million), all with adaptive or AI-driven EW features and results due by 2028.
Thales’ Golden AI analysis tool, labelled by the French Navy and DGA in November 2025, is credited with cutting radar-intercept analysis time by up to four times.
ASELSAN booked a $410 million electronic warfare export to Poland in December 2025, the largest EW deal by a Turkish firm in NATO Europe; its HAVA SOJ jamming aircraft is in pre-delivery flight trials.

What is electronic warfare, and why AI has become the argument

Electronic warfare is the contest for the electromagnetic spectrum. In plain terms it has three parts: listening (electronic support, which finds and identifies radars and radios by their emissions), disrupting (electronic attack, which degrades an opponent’s use of the spectrum) and protecting (electronic protection, which keeps your own sensors and links working). A fighter’s radar warning receiver, a jamming pod and a ship’s decoy launcher are all electronic warfare. None of it is new. What is new is how fast the signals change.

For decades, EW systems worked from libraries. Engineers recorded the parameters of known threat radars in a mission data file, and the aircraft matched what it heard against that list. That model breaks when emitters become software-defined. Modern radars are “increasingly digitized and easily reprogrammable”, as a Thales technical paper published in April 2026 puts it, with modes and waveforms that can be altered by an update rather than a hardware change. A signal that is not in the library may be misclassified or ignored, and the wrong response follows. The same paper notes the scale of the data problem: a radar EW receiver may have to sort “up to tens of millions of pulses” every second, and a communications intercept run can generate several terabytes that a single operator cannot exploit.

Here Europe has a structural problem as well as a technical one. A RUSI paper by Justin Bronk in March 2025 described airborne EW as the area in which NATO’s European members have “the greatest dependence on the US military”, and argued that keeping pace with digital threats “will require increased investment and rapid adoption of AI- and machine learning-enabled toolsets”. Germany’s retirement of the Tornado ECR, Europe’s only dedicated fighter-based suppression aircraft, made the gap concrete. The answer that emerged in Berlin, Linköping and Munich was to pair a new sensor suite with machine learning, and that is the story of the Eurofighter EK.

What cognitive EW actually does, and what it does not

The term comes from the United States. DARPA’s BLADE programme, launched in 2010, set out to use machine learning to detect and characterise new radio threats, synthesise countermeasures in the field and assess whether they worked, replacing what the agency calls the “manual-intensive laboratory-based” approach. Its radar-focused sibling, Adaptive Radar Countermeasures, followed in 2012 with BAE Systems as prime. BAE Systems’ own definition of cognitive EW is simply the use of AI or machine learning “to enhance development and operation” of EW technologies, with the caveat that the best implementations still combine automation with human expertise.

In practice the AI does three high-level jobs. The first is classification: sorting a dense stream of intercepts into known emitters, variants of known emitters and genuinely new ones, using what Thales describes as “novelty detection, clustering, learning” so that out-of-catalogue signals are flagged rather than dropped. The second is spectrum awareness, fusing many sensors into one picture and highlighting what has changed since the last mission. The third is adaptive response, where the system proposes or selects a self-protection measure against an emitter it has never seen, on board and within the timeline of an engagement. Helsing describes Cirra as doing exactly that: classifying unknown air-defence emitters, interpreting their intent and enabling a response while the aircraft is still in the threat area. A fourth, quieter job is operator support: Thales’ Golden AI, which trains on a navy’s archive of radar intercepts and labels new recordings, is credited with analysing R-ESM data “up to 4 times faster” and was demonstrated during the French Navy’s Clemenceau 25 deployment.

What none of the European suppliers claim is a fully autonomous jammer that decides for itself. Every official description reviewed here, from Thales to ELT Group to BAE Systems, keeps a human in or “over” the loop, partly for accountability and partly because EW training data is incomplete by nature: pulses are missed, intercepts are shorter than the waveform, signals overlap. The Thales authors warn that models built without those constraints “often lead to inappropriate performance”. That is why the maturity labels matter. A lab result on recorded data and a qualified module flying on a fighter are very different things, and Europe has far more of the former. For a wider look at where the money and the start-ups are going, see Europe’s Defence AI Companies: Who Is Building What in 2026.

Eurofighter EK: the first European fighter with AI in the loop

Germany chose Saab’s Arexis for the EK in 2023 and Airbus placed the production orders on 14 November 2025: €291 million for the Arexis pods and, subject to approval, a further €258 million that “incorporates AI technology from Saab and strategic partner Helsing”. Saab said deliveries would run from 2025 to 2028. The first step covers 15 Tranche 4 aircraft modified to EK standard, carrying Arexis in wingtip pods alongside the AGM-88E anti-radiation missile, and a further 20 new-build Tranche 5 jets are also to receive the suite, giving roughly 35 Arexis-equipped Eurofighters in all. At ILA Berlin in June 2026 an Airbus spokesman confirmed the 15 kits were in production in Sweden and would be assembled at Saab Nürnberg for integration at Manching “within the next two years”. The Luftwaffe intends to have the capability in operational use from 2030, replacing the Tornado ECR.

The AI itself is Helsing’s Cirra, and the contract signed in Munich in November 2025 gives Helsing three years to embed it in the Arexis sensor chain. Helsing said it had already flown Cirra on a company-owned test aircraft during the 18-month concept phase, and co-founder Gundbert Scherf spoke of moving “into the development phase together with Saab”. Arexis brings the hardware: gallium nitride AESA arrays, ultra-wideband digital receivers and digital radio frequency memory, a combination Saab says is in serial production for Gripen E with onboard AI that needs minimal aircrew attention. The Bundeswehr’s roadmap does not stop there: a second EK step is intended to add an escort jammer in the mid-2030s, according to trade reporting in January 2026. Which pod wins is not publicly confirmed; Hensoldt’s Kalaetron Attack, flight-tested on a PC-12 and shown again at ILA 2026 as an airborne solution for countering enemy air defences, is the obvious German candidate.

Two later moves show how fast the software side is evolving. On 25 September 2026 Saab announced a partnership with Finnish company NestAI to bring “adaptive” AI into selected parts of Arexis, the idea being that the EW system continues to learn after it leaves the factory rather than waiting for the next library update. The Helsing relationship also runs deeper than one contract, as Helsing Explained: Europe’s Defence AI Firm, Drones, Funding sets out. For readers weighing the platform itself, the Gripen E profile and the Saab company profile give the industrial background.

Who is building it in Europe: the company and programme map

The Eurofighter EK is the flagship, not the only line of work. The table lists named systems and programmes for which a recent source describes an AI or adaptive role, with a maturity label for each. The AI column is deliberately high level; official releases rarely go beyond “classification”, “data fusion” or “adaptive”.

System / programme (country)	Lead companies	Platform	AI role (high level)	Status	Latest milestone
Eurofighter EK Step 1, Arexis + Cirra (DE)	Saab, Helsing, Airbus	Eurofighter (15 retrofit + 20 Tranche 5)	Classification of unknown emitters, adaptive self-protection	PROCUREMENT	Kits in production, integration by 2028 (Jun 2026)
Arexis on Gripen E (SE)	Saab	Gripen E/F	Onboard threat classification; adaptive learning planned	SERIAL PRODUCTION	NestAI adaptive-AI partnership (Sep 2026)
Kalaetron Attack (DE)	Hensoldt	Pod, PC-12 testbed; Eurofighter target	“Cognitive elements” for classifying unknown threats	FLIGHT TEST	Shown at ILA Berlin (Jun 2026); no order confirmed
PEGASUS, Kalaetron Integral (DE)	Hensoldt, Lufthansa Technik Defense	3 x Global 6000	Data fusion for SIGINT (not detailed publicly)	PROCUREMENT	Integration complete end-2026, first delivery 2027 (May 2026)
Praetorian eVolution (UK/IT/ES/DE)	Leonardo, ELT Group, Indra, Hensoldt; BAE Systems integrator	Eurofighter Typhoon	Cognitive EW on high-fidelity digital receiver data	DEMONSTRATOR	Component flight trials 2023-24; concept complete (Nov 2024)
Golden AI and CURCO (FR)	Thales	Naval R-ESM analysis; drone-borne radar detector	AI labelling of intercepts, up to 4x faster	FIELD TRIAL	PERSEUS label from French Navy, DGA, AID (Nov 2025)
Tethered-drone EW + IndraMind (ES)	Indra	4×4 vehicle with quadcopter mast	Cognitive layer for data fusion and live translation	PROTOTYPE	Unveiled at Eurosatory (Jun 2026)
REACT II (EU, 7 nations)	Indra-led, 20 partners	Airborne escort / stand-off pods	Reconfigurable EW with digital beamforming	DEMONSTRATOR	OCCAR co-funding contract, €69 million (Feb 2026)
SCEPTER (EU, 9 nations)	Indra, Leonardo, Hensoldt, Thales, Saab + 9	Multifunction RF demonstrator	AI spectrum management, digital twin	LAB	Demonstrator phase under way (May 2026)
AI-WASP (EU, 9 nations)	Patria-led, 16 partners	Crewed and uncrewed aircraft payload	AI-controlled converged comms, RF surveillance and attack	CONCEPT	€45 million EDF award (May 2025)
GCAP ISANKE & ICS (UK/IT/JP)	Leonardo UK, Leonardo, ELT Group, Mitsubishi Electric (G2E)	GCAP fighter	Integrated sensing and non-kinetic effects	CONCEPT	G2E consortium formed (Sep 2025)
HAVA SOJ (TR)	ASELSAN, Turkish Aerospace	4 x Global 6000	“Special algorithms” per Janes; AI not claimed	FLIGHT TEST	Pre-delivery trials shown by MoD (Jun 2026)

Fighters and self-protection

Beyond Germany, the biggest fighter-EW decision is what happens to the Typhoon’s Praetorian defensive aids sub-system in the UK, Italian and Spanish fleets. The EuroDASS consortium of Leonardo, ELT Group, Indra and Hensoldt, with BAE Systems as integrator, finished the Praetorian eVolution concept and flew a digital receiver and band-extension hardware on a Typhoon in 2023 and 2024. Leonardo’s description states that cognitive EW “using AI and machine learning will exploit the high-fidelity data captured and respond to new threats as they emerge”, and that the retrofit needs no airframe changes. It remains a demonstrator; no partner nation has announced a production contract. The same industrial cluster carries forward into GCAP, where the GCAP Electronics Evolution consortium formed in September 2025 (Leonardo UK, Leonardo, ELT Group and Mitsubishi Electric) is responsible for the integrated sensing and non-kinetic effects system, as explained in the GCAP primer.

Elettronica, trading as ELT Group, is the specialist in that group. It closed 2025 with €700 million of orders, up 47 per cent, on revenue of about €370 million, and raised R&D from €50 million to around €70 million in a year. Chief executive Domitilla Benigni says the company does “not want to compete with AI developers” but to use the best of them; its published AI applications so far are in EMSO planning, simulation and the KARMA counter-drone system, not the jammer itself. Hensoldt’s Kalaetron Integral, meanwhile, will fly on Germany’s three PEGASUS signals-intelligence jets, with integration due to finish at the end of 2026 and first delivery in 2027.

Land, naval and tactical EW

The picture is less tidy on the ground, where Ukraine has driven demand towards small, networked systems. Indra’s Eurosatory 2026 exhibit was a quadcopter tethered to a 4×4 that lifts communications-band EW sensors more than 100 metres, feeding its IndraMind “cognitive layer”, which fuses data and translates intercepted voice in real time so that any soldier can use it without specialist training. The Spanish Army and Navy are launch customers. Indra also leads two of the three EU projects in the table. Rohde & Schwarz, with €3.16 billion of revenue, positions itself on “AI-enhanced data fusion” for SIGINT and markets Ukrainian firm Infozahyst’s EW systems abroad. In Germany, the Bundeskartellamt cleared an EW joint venture of Hensoldt, Rohde & Schwarz, Plath and General Dynamics on 28 September 2026 for a single Bundeswehr programme, a sign of consolidation around a national bid.

The EU programmes

Here the money tells the story. The European Commission’s December 2025 factsheet lists four EDF electronic-warfare projects due to deliver between 2025 and 2028: REACT II (€40 million EU funding) for reconfigurable airborne electronic attack, SCEPTER (€35 million) for a multifunction radar-EW-communications aperture with “adaptive cognitive features”, AI-WASP (€45 million) for an AI-controlled swarm payload, and CARMENTA PF (€32.9 million) for a next-generation aircraft self-protection system. OCCAR signed the REACT II co-funding contract in February 2026 with seven nations and a €69 million total budget. Patria’s AI-WASP, with 16 partners, is the most ambitious on paper: one AI-managed aperture for communications, RF surveillance and electronic attack on crewed and uncrewed aircraft, overlapping with the swarm work in Drone Swarms in Europe: Countries, Companies and Programmes. All three are demonstrators or earlier. None has a customer yet.

Ukraine as the live laboratory, and the US comparison

Ukraine is where the argument for cognitive EW stopped being theoretical. CEPA’s Michael Newton, writing in July 2026, described Ukrainian design cycles “measured in weeks rather than years”, with units building and modifying systems close to the front while Russia “adapts constantly”. A BAE Systems business-development director put the same point in a March 2026 note: a communications channel “that works in the morning could well be jammed by the afternoon or even earlier”. The most visible EW counter-move has not been an algorithm at all but the fibre-optic FPV drone, which trails a cable instead of a radio link. That is a useful corrective: adaptation at the front is mostly hardware, doctrine and tempo, and AI classification tools are arriving through Western partners such as Rohde & Schwarz rather than being born there.

The United States remains the reference point, and the comparison is not flattering for Europe on fielded systems. DARPA’s BLADE and ARC programmes are complete, and Military Aerospace’s December 2025 survey of AI in EW lists Thales, Saab, Leonardo and Hensoldt among the players but names only American programmes of record. Europe’s counter is that its programmes are newer and more openly AI-native: on public evidence the Eurofighter EK is the first fighter in NATO Europe to carry a named deep-learning classifier as a contracted deliverable. Whether that is a lead or a late start depends on the 2028 to 2030 dates being met. The governance debate around machine-speed decisions in the spectrum is part of a wider European conversation covered in Agentic AI in Defence: What European Militaries Are Testing.

That is where the industrial bottleneck starts. Cognitive EW needs three things Europe has in short supply at once: flight-test hours against real threat radars, sovereign training data (Thales’ paper stresses national intercept libraries and “data sovereignty”), and engineers fluent in both RF hardware and machine learning. ELT Group’s plan to hire more than 300 people in 2026 is one answer to that constraint.

Where Türkiye fits

Türkiye enters this story as an exporter rather than a research partner. ASELSAN reported 2025 revenue of TRY 180.4 billion (about $4.11 billion), up 15 per cent in real terms, with more than $2 billion of new export contracts and a backlog of $20.4 billion. The largest of those exports in electronic warfare was a $410 million contract with Poland’s Armament Agency signed on 19 December 2025, a NATO customer buying Turkish EW at scale for the first time. ASELSAN’s land-based KORAL AD, shown in a new version at SAHA 2026 in Istanbul in May, combines electronic support and electronic attack on one mobile platform for long-range detection, identification and classification of hostile radars, and is in service with Turkish forces. ASELSAN’s public materials describe AI explicitly for its GÖKALP drone interceptor, not for KORAL, and this article does not claim otherwise; its laser and microwave counter-drone lines are covered in Counter-Drone Lasers in Europe: Operational or Prototype?.

The airborne piece is HAVA SOJ, a stand-off jammer on four Bombardier Global 6000 airframes developed by ASELSAN and Turkish Aerospace. Test flights were tracked in February 2026, the first image appeared on 1 March, and the Ministry of National Defence showed the aircraft in primer during pre-delivery trials in a video on 1 June 2026. Janes reports that ASELSAN supplies the “critical mission equipment, software, and special algorithms”; an original 2023 delivery target was missed and no new date is confirmed. If delivered on the current trajectory, Türkiye would field a dedicated stand-off jammer before Germany fields the Eurofighter EK, which is why Turkish EW belongs in a European survey even without a verified AI claim behind it.

Frequently asked questions

What is cognitive electronic warfare?

Cognitive electronic warfare is electronic warfare that uses machine learning to handle signals a fixed threat library cannot. Instead of matching intercepts against a stored list, the system learns to classify unknown or modified emitters, builds a live picture of the spectrum and proposes responses in real time, while an operator keeps oversight. The term originated with DARPA programmes in the early 2010s.

Which European fighter will be the first with AI-based electronic warfare?

Germany’s Eurofighter EK. Saab’s Arexis sensor suite with Helsing’s Cirra deep-learning software is on order for 15 retrofitted Tranche 4 jets, with a further 20 Tranche 5 aircraft also due to receive Arexis. Airbus placed the orders in November 2025, deliveries run to 2028 and the Luftwaffe plans operational use from 2030.

Which companies build AI-enabled electronic warfare in Europe?

Saab and Helsing lead on fighters, with Hensoldt, Leonardo, ELT Group, Indra and BAE Systems in the EuroDASS and GCAP teams. Thales fields AI analysis tools for the French Navy, Rohde & Schwarz and Indra work on tactical and SIGINT systems, and ASELSAN is the main non-EU exporter, with a $410 million Polish contract.

Is cognitive EW operational or still experimental?

Mostly experimental in Europe. The Eurofighter EK is in procurement with deliveries due by 2028; Thales’ Golden AI has been demonstrated at sea; Praetorian eVolution, REACT II, SCEPTER and AI-WASP are demonstrators or studies. Saab’s Arexis is in serial production on Gripen E, but the adaptive-learning features announced with NestAI in September 2026 are new.

Why does Ukraine matter for AI in electronic warfare?

Because it showed that emitters and jammers now change in weeks, faster than any library update cycle. Analysts describe Ukrainian design cycles measured in weeks, and Western suppliers such as Rohde & Schwarz now partner with Ukrainian firms. That tempo is the main justification European companies give for adding machine learning to EW.