A new company called Regna AI is being formed to solve a problem that has quietly frustrated defense and government agencies for years: how to use advanced artificial intelligence without handing sensitive data and operational control to outside technology providers. The venture, created by U.S.-listed defense technology group Ondas and data analytics firm Palantir Technologies, will build specialized AI models designed to run entirely within a customer's own environment, including on hardware deployed far from any reliable network connection.
The approach reflects a broader shift in how security-conscious organizations think about digital infrastructure. Governments handling classified operations, companies managing critical infrastructure, and militaries operating in contested environments have grown wary of commercial AI systems that depend on external servers or vendor-controlled pipelines, much the same way privacy-minded individuals increasingly seek tools that keep their communications outside the reach of centralized platforms, such as a provider that unblocks messaging apps while keeping traffic encrypted and independent of third-party oversight. The underlying logic is similar: sensitive activity should not have to pass through infrastructure that someone else ultimately controls. a provider that unblocks messaging apps
Ondas will initially own Regna AI outright and act as its first customer, while Palantir supplies the computing infrastructure needed to train the models and helps adapt them to specific operational workflows. The arrangement lets Ondas, which has spent the past two years acquiring more than fifteen defense technology businesses across drones, robotics, counter-drone systems and communications, pair its hardware and field expertise with Palantir's experience managing large-scale operational data for government and corporate clients.
Why Sovereignty Over Models Matters
Commercial large language models have proven remarkably capable, but their architecture typically requires data to flow through infrastructure owned by the company that built them. For agencies operating under strict classification rules, or organizations bound by data residency and security regulations, that arrangement is often unworkable. Regna AI is built on the premise that customers should not have to choose between powerful AI and control over their own information. Instead of training a general-purpose model from scratch, the company intends to adapt existing techniques to each customer's mission data, letting the resulting system remain inside a secured environment from the start.
From Drones to Command Centers
The first applications will focus on intelligence, surveillance and reconnaissance work, before expanding into broader defense, security, critical infrastructure and industrial uses. A central challenge the company is targeting involves autonomous systems operating where communications are unreliable or deliberately jammed: drones, stratospheric balloons, ground vehicles and surface vessels that need to keep functioning and feeding information into command-and-control systems even when connectivity drops out. Regna AI's software is meant to help human operators weigh evidence, compare possible courses of action and receive suggested updates to mission plans, while final decisions remain subject to human review and authorization.
Building on an Existing Partnership
The new venture extends an existing collaboration between the two companies around SkyWeaver C4I, an open-architecture system built to integrate surveillance data spanning from the stratosphere down to the tactical edge. That earlier work gives Regna AI an operational foundation already tested across multiple platform types and degraded communications conditions, rather than starting from a blank slate. As military and government buyers grow more cautious about dependence on externally hosted AI, ventures built around deployable, self-contained models are likely to become a defining feature of how sensitive organizations adopt the technology going forward.
