The Army's Capability Program Executive for Intelligence and Spectrum Warfare announced the Rapid Electromagnetic Warfare and Signals Intelligence Commercial Solutions Offering — REWSI — at the end of April, and over the past six weeks it has begun to take shape as something more interesting than a procurement vehicle. REWSI is the acquisition reform half of a story whose technology half has been visible for two years. The cognitive electronic warfare conversation has been dominated by the question of whether AI-enabled signal classification, autonomous response generation, and software-defined radio architectures can keep pace with adversaries who change their electronic order of battle in days rather than months. The answer is increasingly yes — at the bench. The unresolved question is whether the United States can buy and field those capabilities on a timeline that makes the technical answer operationally relevant. REWSI is the Army's first serious attempt to say it can.
The structure is deliberately unfamiliar to industry. Rather than a program of record that selects a single vendor, freezes a baseline, and amortizes integration cost across a decade, REWSI operates as a continuously open call that runs through April 2027 and feeds into a curated library of pre-vetted commercial capability. Submissions are technology-typed rather than program-typed: a SIGINT receiver, an EW effector, a spectrum sensing waveform, a signal classification model, an integration framework. Vetting happens once; commanders draw from the library against mission needs rather than initiating a new acquisition cycle each time the threat picture changes. The model is borrowed in spirit from how cyber capability is fielded — assembled from a vetted toolkit against the current adversary posture — and the analogy is intentional. The electromagnetic spectrum has come to resemble the cyber domain in the speed at which capability is rendered obsolete and the cadence at which new capability has to be inserted to maintain effect.
Why the Library Model Matters for Cognitive EW
Cognitive electronic warfare is not a single capability. It is a stack: RF sensing hardware capable of capturing wideband spectrum at sufficient fidelity to feed downstream classifiers; embedded ML inference that can identify and categorize novel emitters without depending on a pre-built threat library; response synthesis that can generate candidate countermeasures within the closed loop of a single engagement; and an open architecture that allows any layer of that stack to be updated without rebuilding the platform. No single vendor builds the entire stack well, and the stack itself changes as the threat environment changes. A program of record that locks a baseline at contract award is structurally mismatched to a capability whose value depends on continuous insertion of new models, new waveforms, and new sensing approaches. The library model assumes that mismatch and builds around it. The Army can vet a sensing front end from one vendor, a classifier from another, and an effector waveform from a third, and recombine them as mission demands.
This is what gives REWSI its operational significance. The Pacific Defense work funded by the Office of Naval Research — an AI/ML-enabled EW sensor-effector capability built on the Ubiquitous Edge distributed software architecture and aligned to SOSA and CMOSS open standards — is the kind of capability the REWSI library is designed to absorb. So is the cognitive EW algorithm work the Air Force funded at Southwest Research Institute, which targets real-time detection and response against unknown adversary radar threats. So is the cognitive spectrum sensing component of DARPA's broader work on computational responses to surveillance systems. None of these are platforms. All of them are stack components that have to be integrated against a host that already exists and a threat that has already shifted. The library structure is what allows that integration to happen at relevant timescales.
The Validation Problem the Library Inherits
A pre-vetted library is only useful if the vetting holds up under operational pressure, and this is where the cognitive EW acquisition shift creates a downstream problem that has not yet been fully grappled with. Traditional EW evaluation infrastructure was built to test deterministic systems against characterized threats. Cognitive EW changes both sides of that equation: the system itself is non-deterministic in the sense that its responses are generated rather than retrieved, and the threats it is tested against are by definition novel — the entire premise is that the threat library does not contain them. Validating that a learned classifier will not misclassify a friendly emitter as hostile, that a generated response waveform will not produce unintended effects on adjacent spectrum users, and that the closed-loop behavior of an integrated stack will degrade gracefully when one component fails or is jammed, requires evaluation methods the Army does not yet possess at scale.
The Army Research Laboratory's Generative Unwanted Activity Recognition and Defense effort is one piece of that infrastructure — a prototype framework for detecting unpredictable AI behavior in autonomous systems before fielding. It is not yet a vetting standard for the REWSI library, but the shape of the work suggests where the standard will need to come from: behavior-event analysis applied to learned EW components, adversarial red-teaming of generated response waveforms, and runtime monitoring that can catch out-of-distribution behavior during operational use. Without that infrastructure, the library risks becoming a fast pipeline for capability that field-fails at the moment it matters most. With it, the library becomes the mechanism by which the United States closes the spectrum operations gap that the past three years of conflict have measured in hours and days. The acquisition reform is necessary. The verification infrastructure that lets the reform produce trustworthy capability is the work that the next eighteen months will determine.
