Shield AI's Autonomy Software Flew a Foreign Drone in Two Months

In March 2026, Shield AI and a European defense company called Destinus ran a two-month experiment in Segovia, Spain. The goal: see whether Shield AI's autonomous flight software, Hivemind, could be bolted onto Destinus's Hornet drone without rebuilding from scratch. It worked. Hivemind took the Hornet through its paces — rerouting around simulated geofenced airspace in real time, without a human touching the controls. The result, announced on Shield AI's company blog, is a proof that the autonomy stack doesn't particularly care what airframe it lives inside.

That's the story the press release tells. The more interesting question is what it means for Shield AI's strategy, and for the defense customers wondering whether they're buying a product or a platform.

Shield AI, founded in 2015 by brothers Ryan and Brandon Tseng alongside Andrew Reiter, spent its first years building the Nova quadcopter — a drone designed to scout building interiors so soldiers didn't walk into ambushes blind. The real asset was never the hardware. It was Hivemind: the sensor-to-decision software layer that takes in infrared, radar, signals intelligence, and satellite data, builds a model of the environment, and flies the aircraft autonomously. In 2021, Shield AI acquired the V-BAT — a tower-mounted surveillance drone capable of 13-hour flights at 18,000 feet — and used it as the proving ground for Hivemind's combat capabilities.

The V-BAT has since logged more than 35 missions in Ukraine, where Russian electronic warfare makes most U.S. drones useless. Shield AI says the V-BAT survived eight months of Ukrainian jamming tests in 2024 and went on to identify more than 200 Russian targets, according to the company's account as reported by Fortune. The data from that deployment is what Shield AI is now using to sell Hivemind as a standalone product — software that can run on other companies' aircraft, not just its own.

"Operational requirements are evolving quickly, and autonomy must be integrated at the same pace," said Christian Gutierrez, vice president of Hivemind Solutions at Shield AI, in the company's blog post announcing the Destinus results. Tim Moser, chief technology officer at Destinus, put a finer point on it: "The modular architecture of Hivemind allowed straightforward integration with our flight control and mission systems."

Destinus is not a household name in English-language tech press, which is part of why this story matters. The company was founded in 2021 by Mikhail Kokorich, a Russian-born entrepreneur who relocated to Switzerland after exiting a previous venture. (A wire summary misidentified the founder as Uriah Thurnheer; current public records consistently attribute the company to Kokorich.) Destinus makes autonomous strike and air-defense systems for European armed forces, including loitering munitions interception — the Hornet that flew with Hivemind is a multi-role platform designed for counter-UAS and layered air defense. The company has also supplied drones to Ukraine's defense forces and acquired the autonomous flight startup Daedalean for $223 million in mid-2025, signaling ambitions beyond traditional hypersonics.

The portability question is the crux. Defense procurement has historically locked customers into integrated hardware-software bundles — you buy the plane, you use the autopilot the plane came with. Hivemind's pitch is different: what if the software is the product? That's the logic behind Shield AI's selection for the U.S. Air Force's Collaborative Combat Aircraft program, where Hivemind was integrated onto Anduril's YFQ-44A Fury drone — and, notably, the aircraft switched between Hivemind and Anduril's own Lattice autonomy stack mid-flight, completing the same test cards with both. That trick, demonstrated in late February 2026, was only possible because both stacks comply with the Defense Department's Autonomous Government Reference Architecture, a standards framework designed to prevent vendor lock-in by making different autonomy stacks interchangeable on compliant hardware.

The YFQ-44A flight was a two-day turnaround from a separate test in which a YFQ-42A launched with an AIM-120 AMRAAM missile. On the autonomy front, Anduril's senior vice president of engineering Jason Levin said the swap was enabled by early implementation of A-GRA on both the aircraft and the mission autonomy software stack. Shield AI, for its part, framed it as a validation of what it calls a "software first approach" to airpower.

That approach is now being tested across a growing list of airframes. Hivemind has now flown on Shield AI's own V-BAT and X-BAT, Airbus's DT25 target drone (three-month integration, announced in September 2025), Anduril's YFQ-44A, General Atomics' MQ-20 Avenger (February 2025, via A-GRA compliance), and now the Destinus Hornet (two months). The cadence is compressing.

Gary Steele became Shield AI's CEO in May 2025, brought in to scale revenue from roughly $300 million in the year ending March 2025 toward $1 billion by the year ending March 2028 — a 70 to 100 percent annual growth target. The Destinus deal, and Hivemind's expanding roster of host platforms, is the product strategy behind that target: sell the software layer, not just the drone.

There is a gap worth naming between what a demo shows and what a fielded system does. The Segovia tests validated integration and real-time rerouting in a controlled campaign. Destinus describes this as "the first phase of a broader effort" to extend autonomy across its platform families, with future phases covering advanced mission planning, terrain-aware flight, and coordinated multi-platform behaviors. None of that is deployed yet. The Ukraine V-BAT numbers — 35 missions, 200 targets — are Shield AI's own figures, shared via the company's account to Fortune, and haven't been independently audited in a way that's available to press. The YFQ-44A switching demonstration is real, but it's a test flight, not operational combat.

The counterargument for defense customers is real too: primes like Lockheed Martin and Boeing have their own autonomy programs, the A-GRA framework is still maturing, and the Defense Department has a long history of awarding contracts that never field systems. Shield AI's $5.6 billion valuation — hot on the heels of Anduril among defense tech startups — prices in a future where the Android model actually wins in defense. That future isn't guaranteed.

What the Destinus demo adds is evidence that the portability thesis is real, not just marketing. Two months on a European hypersonics company's airframe is a different kind of proof than another press release from a company's own test range. Whether that translates into contracts — and whether those contracts translate into systems that operators trust with their lives — is the next question. For now, Hivemind has one more airframe to its name, and Shield AI has one more data point for the pitch deck.