“Cognitive electronic warfare” is about to mean nothing. It should mean a single platform that gets smarter with every mission.
On 26 July 2026, a Ukrainian radio specialist tracking Russian FPV drones noted a key detail: the Boomerang drone had shifted its video channel above 6 GHz. Not a doctrine update. Not a procurement decision. A frequency shift, made in the field, in response to whatever was jamming it that week.
This is the environment that “cognitive electronic warfare” was meant to describe. A threat that adapts faster than a requirements document can be written, let alone approved.
A term that now describes everyone
Across the industry, the same phrase now appears on nearly every serious vendor's page, and for good reason: the whole sector is genuinely investing in this direction, from established primes to specialist sensing firms and research partnerships. Individual components and full architectures alike are increasingly grouped under the same label as the technology matures towards it. Market analysts now use the term as a category header spanning airborne jammers, naval sensor suites, and reprogrammable self-protection systems across a dozen companies, each pursuing a different architecture towards the same goal.
When a term expands to cover an entire market, it ceases to describe anything specific. That is not a semantic complaint. It has practical consequences for the people whose job is to evaluate these systems before they get anywhere near a mission. A procurement officer reading “cognitive” on a data sheet has no way of knowing whether it refers to a single component with a learning algorithm bolted on or to an architecture built around continuous adaptation from the ground up.
This pattern is not unique to defence, and it will not be confined to marketing copy. Even within the aerospace and defence industry, confidence is not universal: a 2026 industry outlook found that while executive leadership is broadly optimistic about artificial intelligence, middle management, the people who actually have to operate and integrate these systems, remains considerably more sceptical of claims that outpace demonstrated results. That gap is worth taking seriously, because it is operators, not boardrooms, who decide whether a system earns trust in the field. Inflated claims eventually face an operational test. The gap becomes visible, and scepticism spreads to genuine capability alongside the false one. Defence procurement, where the cost of being wrong is measured in mission outcomes rather than market share, has less room for that pattern than most industries.
What the term should mean
Set the marketing language aside. The research is clear about what distinguishes cognitive electronic warfare from the systems it replaces. NATO's own science and technology work on the subject is explicit: modern adaptive radar and radio systems continuously change modulation, bandwidth, frequency, and timing, and conventional signal intelligence methods that rely on matching a fixed reference library cannot keep pace. The research task groups studying this problem are not asking how to add a machine-learning feature to an existing jammer. They are asking how a system builds and updates a working model of the electromagnetic environment as it operates, engagement by engagement.
This is not a niche concern within a single research programme. A peer-reviewed analysis of NATO doctrine published this year reached a similar conclusion at the alliance-strategy level, noting that member states are investing heavily in autonomous and adaptive systems without yet sharing a framework to integrate them. The same gap appears at the unit level: an analysis of electronic warfare readiness among NATO brigades in the Baltic region, published this month, found that many tactical radio networks still rely on largely static frequency plans, precisely the configuration a fast-adapting threat is built to exploit. The problem repeats at every level, from doctrine down to a single radio net, because it is the same problem: systems designed around what was known are being applied to threats that no longer hold still.
That is the real bar. A cognitive electronic warfare platform does not rely on a threat library updated on a schedule. It builds a model of the environment from what it has already observed, combining RF, acoustic, and optical inputs. That model sharpens with every engagement, not with every software release. It produces a threat assessment with an attached confidence level, not a binary alert. And it recommends a course of action rather than executing one, because the platform's job is to compress decision time, not to remove the decision.
That last point separates a serious cognitive EW claim from a marketing one just as clearly as the technical architecture does. A system that acts on its own, without an operator approving the response, is not solving the problem that NATO's research groups are working on. It is creating a different one.
“The industry has spent two years slapping the label ‘cognitive’ on anything with a chip in it. We’d rather be measured against what the term is supposed to mean: a platform that knows more after every mission than it did before, and still waits for the operator to decide.” Jari Mielonen, CEO, Crown Defence
What this means for Onyks™
This is the standard Crown Defence-built Onyks™, which is why the platform is architected around a single cognitive core rather than a RF unit with a learning add-on. RF, acoustic, and optical inputs feed the same model. Every engagement, successful or not, updates the model. The operator reviews the recommendation and issues the order; the system executes only once authorised.
Onyks™ has met this standard through operational testing, not in laboratory conditions. Development has been shaped directly by user feedback and the platform is field-tested rather than a claim made from a specification sheet.
The frequency shift introduced by a Ukrainian FPV drone in July did not wait for anyone's next product cycle. Neither should the systems built to counter it.
