Adrenum is building distributed sonar sensing to detect autonomous narco subs and modern maritime threats

Summary

Matt Cernek, CEO and co-founder of Adrenum, is building distributed sonar sensing systems designed to fill a critical gap in maritime domain awareness. The company, incorporated in June 2025 and headquartered in Los Angeles after bootstrapping from a garage in Colorado, positions itself as the undersea equivalent of low-earth-orbit satellite constellations — distributing sensing coverage at scale rather than relying on a small number of high-cost exquisite systems.

The core thesis is that legacy sonar infrastructure, built during the Cold War for tracking Russian submarines across the Atlantic, is fundamentally ill-suited to today's threat environment. Autonomous underwater drones, narco semi-submersibles, and asymmetric maritime threats now dominate the risk landscape, and existing systems were never designed to detect them. Cernek points to Ukraine's successful use of underwater drones against the Kerch (Crimea) bridge as a concrete illustration of the capability gap.

Counter-narcotics is an immediate commercial vector. Cernek notes that border apprehensions on land are sharply down, pushing cartel activity into maritime corridors — drug smuggling, human trafficking, and autonomous narco boats, including one recently reported to have been equipped with Starlink. The cartels' shift to crewless vessels, he argues, is driven by the risk that human crew members will talk. Semi-submersible narco boats typically run diesel or outboard engines and are acoustically detectable at significant range. Adrenum's pitch to DHS — whose budget was increased substantially in the recent reconciliation legislation — is to deploy sensing nodes at known maritime choke points to track, pattern, and ultimately interdict these routes.

The product architecture combines proprietary hardware and a vertically integrated software stack, from the physical sensor through digital signal processing to cloud-based perception models. Cernek argues hardware ownership is non-negotiable: accessing data from legacy contractor systems requires navigating layers of government procurement bureaucracy, making it faster and cleaner to build from scratch. The AI/ML layer is being developed by a former early Androl perception engineer — described as their seventh hire — who views sonar classification as a substantially harder problem than the now-commoditized computer vision work done for aerial drones.

The foundational data strategy mirrors how vision models were trained: Adrenum is building what it describes as the world's largest sonar database, initially classifying signals into two buckets — biologics (whale calls, snapping shrimp, ambient ocean noise) and man-made signatures — then progressively refining to vessel type, propulsion configuration, and AIS correlation. At a Navy demo last summer, a former sonar technician using Adrenum's UI identified an 8-cylinder diesel tugboat across a harbor, including its speed and propeller count, from the acoustic signature alone. The near-term product output to end users is a tip-and-cue interface: an alert that a man-made object without AIS is present, with classification detail improving as the training dataset grows.

Cernek is recruiting retired submarine sonar technicians to accelerate classification development, using their pattern recognition expertise to label training data at the granular level that automated systems cannot yet replicate independently. The analogy he draws to radiology — AI as co-pilot surfacing anomalies for human review — reflects the current product maturity stage, with the expectation that autonomy increases as deployment scale expands.