Breaking News
Leonardo to equip Proteus uncrewed helicopter with Shield AI Hivemind for autonomous maritime missions.
Leonardo signed an agreement with Shield AI on October 29, 2026, to integrate the Hivemind mission autonomy Software Development Kit (SDK) into its Proteus uncrewed helicopter demonstrator. The partnership enables Leonardo engineers to build, retain, and adapt customized AI decision-making software for Royal Navy anti-submarine warfare, ISR, and contested logistics operations while maintaining proprietary IP ownership. Developed under a £60 million UK Ministry of Defence programme, the three-tonne aircraft completed its maiden flight on January 16, 2026, and features a 1,000 kg modular payload compartment built for two NATO 463L pallets.
Leonardo will integrate Shield AI's Hivemind SDK into the British Royal Navy's Proteus uncrewed helicopter to develop autonomous anti-submarine warfare, intelligence, surveillance, reconnaissance, and contested logistics capabilities. (Picture source: Leonardo)
On October 29, 2026, Leonardo signed an agreement with Shield AI to integrate the Hivemind Software Development Kit (SDK) into the Proteus, a three-tonne-class uncrewed helicopter intended for the British Royal Navy's future anti-submarine warfare (ASW), intelligence, surveillance and reconnaissance (ISR), and logistics missions. Developed under the UK's Rotary Wing Uncrewed Air System (RWUAS) Phase 3B programme, following a Ministry of Defence contract worth approximately £60 million in 2022, the Proteus first flew at Predannack, Cornwall, on January 16, 2026, using Leonardo's own autonomous flight controls. The Hivemind software will therefore support mission decision-making for an aircraft that has already demonstrated uncrewed flight capabilities.
The helicopter can carry approximately 1,000 kg of equipment or two NATO 463L pallets, while Leonardo will retain ownership of the autonomy functions developed through the SDK. No agreement value, integration deadline, Hivemind flight-test date, or Royal Navy production order has been announced. An initial prototype of the Proteus was displayed at DSEI 2023 before completing synthetic mission demonstrations and conducting its maiden flight in January 2026, monitored by test pilots on the ground. Its modular compartment can accommodate two NATO 463L pallets, each measuring 2.235 × 2.743 metres, although usable cargo dimensions and load-distribution limits remain undisclosed.
To date, Leonardo examined 16 mission sets, including ASW with sonobuoys, anti-surface warfare, radar-based airborne early warning, ISR, search and rescue, and intra-theater transport between naval units. These configurations cannot be assessed solely by payload mass because available electrical generation, cooling capacity, payload placement, vibration limits, and flight endurance also determine mission feasibility. The Proteus demonstrator also contributes to the Royal Navy's Maritime Aviation Transformation strategy through 2040, but its speed, endurance, operational radius, service ceiling and shipboard operating limits remain undisclosed. No procurement quantity or replacement schedule for Merlin HM2 or Wildcat HMA2 helicopters has been established so far. The Hivemind's integration likely concerns the mission-management functions above the Proteus' existing autonomous flight controls.
How Shield AI's Hivemind could expand the Proteus helicopter's autonomy
On other aircraft, the Hivemind has issued heading, speed and altitude commands, followed waypoints, avoided designated geographic restrictions, modified routes in response to detected obstacles and coordinated manoeuvres with other autonomous agents. Leonardo engineers will therefore develop their own applications using Shield AI's Software Development Kit (SDK), retaining the resulting intellectual property while developing and validating additional behaviours. Processor specifications, sensor-processing latency, command-update frequencies and redundancy arrangements remain undisclosed. For instance, an autonomous ASW mission would require more than flying a search pattern: the unmanned helicopter would need to manage sonobuoy deployment, receive acoustic information, maintain contacts, adjust its search activity and transmit the submarine contact information.
These activities also require the aircraft to distinguish between decisions it can make independently and actions requiring operator authorisation, particularly when mission equipment or weapons are involved. The Hivemind development elsewhere has used synthetic simulation, software-in-the-loop validation, hardware-in-the-loop testing and supervised aircraft trials, but none of these stages has been confirmed as completed for the Proteus integration. The Airbus H145 provides the closest rotary-wing comparison through the U.S. Marine Corps Aerial Logistics Connector (ALC) programme, which involves Shield AI, Airbus U.S. Space & Defense, L3Harris Technologies and Parry Labs. During the fourth flight-test period, all four companies' systems operated together for the first time, with the Hivemind software running on Parry Labs computers, receiving perception information through L3Harris avionics and commanding the H145's existing four-axis autopilot.
During point-to-point logistics trials, the helicopter detected obstacles in its intended landing area and diverted to another landing zone without crew commands. Testing also varied headings, altitudes, approach distances and obstacle positions, progressing from a large SUV at the landing-zone centre to smaller objects near the touchdown area. The exercise tested the connection between perception, mission decisions and flight-path execution, including the ability to reject a landing area after the approach had begun. The trials involved prepared surfaces, however, and did not establish performance on unprepared terrain, during external sling-load operations or under degraded GPS and communications conditions, but established a precedent for connecting the Hivemind to an existing helicopter's autopilot and third-party perception equipment. The MQ-20 Avenger and X-62A VISTA programmes tested Hivemind under substantially different flight-control and combat conditions.
From the H145 to the F-16 - How Hivemind has been tested in flight
Integration on General Atomics' MQ-20 began before Thanksgiving 2024, leaving 12 weeks before the first flight despite governmental safety restrictions on access to the aircraft's dynamics simulator; engineers used a substitute model before adjusting the software to the actual simulator within one day. At Orange Flag 25-1 in February 2025, the Hivemind demonstrated A-GRA-compliant command-and-control interfaces, heading, speed, and altitude inputs, waypoint navigation, and geozone avoidance. Four months later, one physical MQ-20 and its digital twin executed a 180-degree offset combat air patrol, maintaining a one-nautical-mile (1.852 km) trail formation through geometric positioning despite limited speed-control authority. The X-62A VISTA, a modified F-16 experimental jet, instead conducted AI-controlled offensive and defensive engagements against a crewed F-16, including a flight observed by U.S. Air Force Secretary Frank Kendall in 2024.
The agents trained through millions of simulated dogfights per day, with evaluation criteria including an offensive kill rate above 90% and defensive losses below 10% in specified simulated engagements. Live flight testing imposed a minimum altitude of 10,000 feet and separation of at least 1,000 feet between aircraft, while pilot feedback revealed excessive closure rates during some gun approaches, prompting changes to the agents' training constraints to produce more controlled engagements. Other integrations examined whether the Hivemind could coordinate multiple physical aircraft, operate through government-defined software interfaces, and transfer control between competing autonomy systems. In December 2025, two physical Kratos BQM-177A drones conducted coordinated defensive combat air patrols at Point Mugu, California, against simulated adversaries, alongside a virtual F/A-18.
On February 26, 2026, the Hivemind flew aboard Anduril's YFQ-44A Fury in the Mojave Desert, demonstrating initial mission behaviours and mid-mission updates under the U.S. Air Force Collaborative Combat Aircraft programme. Its A-GRA approach separates mission autonomy selection from aircraft procurement, allowing different software suppliers to compete for integration. On March 19, 2026, Northrop Grumman's Talon IQ, based on the Scaled Composites Model 437, executed combat air patrol and target-engagement manoeuvres under Hivemind before transferring control to Northrop Grumman's Prism software, following one day of hardware-in-the-loop testing. Mitsubishi Heavy Industries also integrated Hivemind into its ARMD experimental aircraft programme in fewer than 60 days during March 2026, demonstrating reinforcement-learning manoeuvres against a simulated airborne object using two aircraft types.
How Hivemind expanded from combat aircraft to ships and drones
The BQM-177A trials therefore established coordination between two physical aircraft, the YFQ-44A demonstrated initial mission autonomy on a collaborative combat aircraft, Talon IQ demonstrated software substitution during a flight campaign, and ARMD demonstrated integration with Japanese experimental aircraft. Hivemind has also been tested on unmanned vessels, interceptor drones, and spacecraft. In late October 2025, HII integrated Hivemind Enterprise with its Odyssey autonomy suite aboard a Romulus 20 unmanned surface vessel during three days of testing at Virginia Beach. In September 2026, two Kraken K3 SCOUT vessels coordinated maritime ISR at Portsmouth, using radar, cameras, and AIS information to detect, classify, track, and escort a vessel. Thunder Tiger's SeaShark 600 and 800 completed integration within six weeks by July 2026, exchanging radar and AIS tracks before following a contact and resuming patrols.
On June 15, 2026, Shield AI and Destinus announced Hornet interceptor trials in Segovia, Spain, demonstrating terrain following, in-flight target updates and terminal manoeuvres after fewer than two months of integration, with subsequent application planned for the Ruta drone. Honeywell's October 8, 2026, demonstration near Phoenix involved one aircraft detecting a target and transferring coordinates to another, which autonomously approached and delivered a payload; its earlier BFD-SE8 first flew autonomously approximately six weeks after software training began. An orbital experiment announced on August 24, 2026, involved one physical satellite and five simulated satellites, generating 189 Hivemind commands over 24 hours, averaging 7.875 per hour, subject to Sedaro approval rather than unrestricted spacecraft control.
Consequently, by late October 2026, Shield AI counted more than 40 systems piloted by Hivemind, but integration durations and demonstrated capabilities varied substantially. The SeaShark vessels required six weeks, Mitsubishi's ARMD fewer than 60 days, Destinus' Hornet fewer than two months, and the MQ-20 12 weeks, while Talon IQ's one-day figure covered hardware-in-the-loop testing rather than total development. Operational complexity also differed: the MQ-20 demonstrated navigation and physical-digital coordination, the H145 made sensor-driven landing decisions, two BQM-177As coordinated defensive manoeuvres, and Talon IQ transferred control between competing autonomy systems.
For the Proteus, the unresolved requirements are therefore plausibly more specific: connecting Hivemind to maritime sensors, processing acoustic contacts, managing sonobuoy deployment, maintaining submarine tracks or adapting searches without continuous operator commands. Shipboard recovery adds moving-deck navigation, deck-motion compensation and restricted landing-area requirements, while unknown endurance, electrical generation and mission-equipment weights prevent calculation of sonobuoy capacity, patrol coverage or sortie output. The October 2026 agreement establishes none of these capabilities. The next measurable steps are Hivemind-controlled Proteus flight, integrated maritime sensor trials, complete autonomous naval mission demonstrations and a Royal Navy procurement decision.
Explore More Defense News
• Land Defense News
• Naval Defense News
• Defense Aerospace News
Written by Jérôme Brahy
Jérôme Brahy is a defense analyst and documentalist at Army Recognition. He specializes in naval modernization, aviation, drones, armored vehicles, and artillery, with a focus on strategic developments in the United States, China, Ukraine, Russia, South Korea, Türkiye, and Belgium. His analyses go beyond the facts, providing context, identifying key actors, and explaining why defense news matters on a global scale.