A new company called Regna AI is being formed to solve a problem that has quietly vexed defense planners and government technologists for years: how to bring frontier artificial intelligence into classified and mission-critical operations without handing sensitive data, infrastructure or decision-making authority to outside commercial providers. The venture, announced jointly by defense technology firm Ondas and data analytics giant Palantir Technologies, will build specialized AI models designed to operate inside customers' own secure environments rather than in the cloud systems that power most commercial AI products.
Ondas, a U.S.-listed company with a substantial engineering workforce in Israel, will initially own Regna AI outright and act as its first customer, while Palantir contributes the computing infrastructure needed to train and refine the models. The arrangement reflects a broader shift in how security-conscious organizations think about digital tools more generally - the same instinct that leads privacy-minded individuals to control where their data travels and who can see it is now driving institutional decisions about AI. Just as a consumer might consult a guide to setting up a VPN on an iPhone to keep personal communications out of reach of third parties, governments and militaries are increasingly demanding that their AI systems keep operational data entirely within environments they control. a guide to setting up a VPN on an iPhone
Why Off-the-Shelf AI Falls Short for Sensitive Work
Commercial large language models have demonstrated real capability, but their architecture typically assumes data will flow to centralized servers operated by the model's developer. For a bank, that might be an acceptable trade-off. For a military unit conducting intelligence, surveillance and reconnaissance, or a government agency managing critical infrastructure, it is often disqualifying. Classified networks, export-control regimes and operational security rules generally prohibit sending sensitive data outside tightly controlled boundaries, which has left many such organizations unable to use the most capable AI tools without first stripping out the context that makes them useful.
Regna AI is built around the idea that customers should not have to choose between capability and control. Rather than building enormous general-purpose models from the ground up - a costly and resource-intensive undertaking - the company intends to develop narrower, mission-specific models that can be deployed directly onto hardware in the field, including drones, ground vehicles, surface vessels and even stratospheric platforms, as well as into command-and-control systems used to coordinate them.
Operating Where Connectivity Cannot Be Guaranteed
A central use case for the new venture involves autonomous systems operating in environments where network connections are unreliable, intermittent or actively jammed by an adversary. In such conditions, AI models that depend on constant contact with a remote server are of limited use. Models that can reason locally, on the device itself, offer a way to keep systems functional even when communications links fail.
The software is intended to help human operators weigh evidence, compare possible courses of action and receive suggested updates to mission plans, with final authorization remaining a human responsibility. That emphasis on human oversight reflects ongoing debate within defense and policy circles over how much autonomy should be delegated to machines in operations with life-and-death consequences.
Building on Existing Infrastructure
Regna AI will draw on SkyWeaver C4I, an open-architecture surveillance and command system already developed jointly by Ondas and Palantir, which integrates data from the stratosphere down to the tactical edge. That existing collaboration gives the new venture an operational base to build from rather than starting entirely from zero. The launch also follows a rapid run of acquisitions by Ondas, which last month bought three Israeli defense technology firms for a combined $56 million, bringing its total portfolio to more than 15 businesses spanning drones, robotics, counter-drone systems and communications assembled over roughly two years.
