Leonardo Bought the Autonomy, Not the Autopilot

In a hangar at Yeovil, a three-tonne helicopter with no cockpit is being wired for a second brain.

The aircraft is Proteus, Leonardo's uncrewed rotorcraft demonstrator for the Royal Navy. On 29 October, Shield AI said Leonardo will use its Hivemind software development kit to write new mission autonomy for it: anti-submarine warfare, surveillance, and logistics flights into places where satellite links and radio are unreliable.

No price. No duration. No delivery date. The announcement ran five paragraphs and two quotes.

That absence is the story. The airframe was already flying. What changed is who writes the decisions.

Proteus first flew on 16 January 2026 at Predannack in Cornwall. The autonomy on board was Leonardo's own. Hivemind arrives as an upgrade layer, not as the thing that lifted it off the ground. Read the coverage quickly and you can get that backwards.

The toolkit, not the helicopter

Shield AI is selling access to an engine, not a finished pilot. Leonardo's engineers get the SDK, a simulation environment, and baseline autonomy algorithms. They write the mission code on top: the acoustic search patterns, the route logic, the rules for handing control back. Shield keeps the core. Leonardo keeps what it builds.

"Autonomy software is never finished," said Tom Schaefer, Shield AI's vice president of Hivemind engineering. "It changes with the mission, the threat, and the sensors on the aircraft."

That sentence explains the commercial model better than the press release does. If the software is never finished, the vendor who owns the base layer gets paid every time a customer changes the top layer.

It also explains why the announcement is thin on numbers. A toolkit has no unit price. It has a licence, a support contract, and a renewal date, and none of those were published.

What Proteus actually is

Proteus is a technology demonstrator, not a procurement programme. It sits inside the UK's Rotary Wing Uncrewed Air System effort, Phase 3B, run by Leonardo with the Royal Navy and UK Defence Innovation. The design is a three-tonne-class vertical take-off aircraft with a modular bay sized for about one tonne of payload or two standard NATO 463L pallets.

That payload figure matters more than the aircraft's speed. A bay that takes two cargo pallets can carry sonobuoys, radar, or resupply: three different missions on one airframe. Leonardo was tasked with assessing 16 mission sets for the demonstrator. The Royal Navy frames it as part of a hybrid air wing that runs uncrewed aircraft alongside crewed ones through 2040.

The money already committed is separate. Leonardo won a four-year, £60 million contract for the earlier Phase 3A demonstrator work in July 2022. The new software agreement has no disclosed value. Conflating the two is the most common error in the coverage so far.

What the aircraft demonstrates is a design bet: keep the airframe modular, keep the payload swappable, and let software decide what the machine becomes on any given day.

Four primes, one engine

The Leonardo deal is the fourth agreement of this kind in under two years. The table is the argument.

Company announcements, March 2025 to October 2026

Honeywell is the proof point. It started Hivemind training in the summer and flew a two-aircraft ISR mission in October. The company estimated the work would have taken 54 weeks from scratch. It took six.

Shield AI says Hivemind has now piloted more than 40 platforms: F-16s, jet-powered uncrewed aircraft, helicopters, surface vessels, ground vehicles, and a satellite in low Earth orbit. That breadth is the product. A prime does not buy an autopilot for one aircraft. It buys a stack that already knows how to fly several.

Leonardo builds some of the most widely flown rotorcraft in the world, and Hivemind gives its engineers new tools to add autonomy to them.— Tom Schaefer, vice president of Hivemind engineering, Shield AI

Why a prime rents the brain

Three forces push a company like Leonardo toward a licence instead of a build.

The first is time. Autonomy teams are scarce and expensive, and a demonstrator programme has a date attached. Buying a working SDK removes two years of foundation work before the first useful test.

The second is ownership. Under the Hivemind model, the partner keeps the mission code it writes and can maintain it without the vendor. For a prime selling sustainment contracts that run for decades, that is not a detail. It is the deal.

The third is portfolio reach. Adam Wardrope, Leonardo's vice president for helicopter market development in the UK, said the company could extend Hivemind to other programmes in its rotary-wing line. One licence, many airframes.

The counterweight is obvious. The base layer stays with the vendor. Every future capability has to be built on someone else's foundation, and the price of that foundation is set by someone else.

CHRONICLE: Hivemind goes horizontal
────────────────────────────
MAR 2025  ·  KAI licenses Hivemind for AI pilots
JAN 2026  ·  Proteus first flight on Leonardo autonomy
JUL 2026  ·  Honeywell signs Hivemind Enterprise MoU
OCT 2026  ·  Honeywell flies two-aircraft ISR in six weeks
OCT 2026  ·  Qatar MoU, then the Leonardo SDK deal
◉ NOW     ·  Four primes, one base layer

The part nobody priced

None of that makes the deal weak. It makes it unmeasured. The useful question is not whether Leonardo can build autonomy on Hivemind. It is what Leonardo can still do if the licence terms change, the vendor is acquired, or an export rule moves.

As we wrote in our November piece on the Navy's $40 million robot warship, autonomy at sea already has a visible price tag. This agreement does not. The gap between those two numbers is where the next round of procurement arguments will happen.

What to watch

Three markers will tell you whether this is a demo or a doctrine.

First, Hivemind-authored mission code actually flying on Proteus. The SDK is installed; the flight test is not scheduled in public.

Second, whether Leonardo extends the licence beyond one demonstrator. Wardrope's line about other programmes is a signal, not a commitment.

Third, whether other European primes copy the structure. A licensing model spreads when the second and third customers get the same terms as the first.

The airframe was never the moat. The decision layer is.