From a commercial call to production contracts
REWSI is the Rapid Electromagnetic Warfare and Signals Intelligence Call for Solutions under the Army Open Solicitation and Commercial Solutions Opening framework. The Army describes rolling evaluations during the open call and an effort to establish a collection of commercial capabilities. That supports a more responsive acquisition model; it does not remove the need for contract actions, integration work or authorization for a particular operational use.
September 2026 update: The Army announced its first two REWSI production contracts on September 10. Heaviside Industries received a $99 million IDIQ for the MOTH spectrum analyzer, and Research Innovations received a $96.9 million IDIQ supporting Dragonfly sensors on the Menet Aero Intrepid 1200 unmanned aircraft. Both have five-year ordering periods. Those contract values describe ordering capacity, not proof that the full amounts have been spent or that every capability has been fielded.
The acquisition idea is consequential. A formation may need a different receiver, classifier or spectrum-awareness capability as its operating environment changes. A reusable collection can make those options easier to find and evaluate. Its long-term value depends on how much tested integration work carries forward to the next mission and how clearly new work is identified.
Cognitive EW is a stack of capabilities
Electronic warfare that adapts to unfamiliar signals draws on several functions. Keeping them distinct makes both proposals and evaluations more useful:
- Sensing: Capture signals with adequate coverage, timing, sensitivity and context for the intended environment.
- Interpretation: Extract features, associate observations and classify emitters while exposing uncertainty about unfamiliar or ambiguous signals.
- Decision support: Recommend or select an authorized response within mission constraints.
- Execution: Apply the response through the appropriate hardware and interfaces, then assess what happened.
- Integration: Exchange data and commands across components without losing provenance, timing or control of the configuration.
Different suppliers may contribute different parts of that stack. Open interfaces can widen the available choices, but an interface specification alone does not demonstrate that components work together under load. Buyers still need compatible data meanings, timing behavior and failure handling.
There is an established research base behind this approach. Pacific Defense's announcement of a $17.4 million Office of Naval Research effort describes AI/ML-enabled electromagnetic warfare using ONR’s Ubiquitous Edge architecture and CMOSS/SOSA-aligned systems. Southwest Research Institute's April 2024 announcement describes a $6.4 million Air Force research contract to improve recognition of unfamiliar radar signals, including feature extraction and neuromorphic computing. These are supplier descriptions of funded development, not evidence of universal operational performance.
Their relevance is the combination of adaptable software and modular hardware. A new classifier is more valuable when the receiving platform can accept it without rebuilding the entire system. The same modularity also makes configuration control more important: the buyer must know which sensor, software and model combination produced a test result.
Speed depends on reusable evidence
A familiar emitter in a clean laboratory environment is only one test case. A useful evaluation should also cover overlapping friendly and adversary activity, weak or incomplete observations, unexpected operating modes and changes in the surrounding spectrum.
Misclassification can have consequences beyond a poor label on an operator display. A response based on that label may interfere with friendly communications or consume a limited operational opportunity. That is why the acceptable uncertainty for a situational-awareness aid may differ from the uncertainty acceptable for an automatically initiated effect.
Programs should ask for evidence about:
- Boundaries: Which signal types, environments and platform configurations were tested, and which remain outside the demonstrated scope?
- Unknowns: How does the system identify insufficient evidence rather than forcing an unfamiliar signal into a known category?
- Mission permissions: Who can approve changes in response authority, and how are those permissions enforced locally?
- Friendly-system effects: What coexistence testing covers adjacent systems and intended operating conditions?
- Degraded operation: What remains available when sensing, communications or compute capacity falls below the expected level?
- Updates: Which regression tests are repeated after a model, waveform, firmware or interface change?
These questions apply whether an implementation is deterministic, probabilistic or a combination. The test obligation follows the behavior and its consequences, rather than the presence of an AI label.
What the GUARD research effort contributes
The Army's January 2026 GUARD prototype award provides a related assurance example. The public award notice describes work directed by the Asymmetric Warfare Division, administered through Army Contracting Command–Orlando's TReX II arrangement with Advanced Technology International and Battelle. Its proposed Behavior-Event Graph approach combines explainability and other AI methods to examine difficult system behavior and produce risk profiles.
GUARD is a research approach to assurance, not a published REWSI approval standard. Its useful connection is the need to examine behavior across sequences of events. A classifier that looks acceptable in isolation may still contribute to an unsafe chain when another component treats a tentative result as an instruction.
For an acquisition team, the practical deliverable is a repeatable evidence package: interfaces, configuration identifiers, scenario coverage, measured results, known limits and the actions required after a change. That package lets a buyer reuse what has actually been demonstrated while focusing new tests on the differences that matter.
Sources and further reading
- Army: first REWSI production contracts and rolling commercial evaluations
- Pacific Defense: ONR AI/ML-enabled EW development announcement
- Southwest Research Institute: cognitive EW research contract
- Army GUARD prototype award notice
Spartan X brings AI, cybersecurity and engineering together around this acquisition problem: making adaptable capability easier to integrate, evaluate and sustain as the mission and spectrum change.
