Two Taiwanese unmanned surface vessels (USV) running Shield AI’s Hivemind autonomy software detected a target vessel and escorted it out of a designated search area off southern Taiwan on 29 July, in what the companies describe as a first on water for multi-asset autonomous teaming.
The demonstration took place at Dapeng Bay in Pingtung County, with personnel from Taiwan’s National Chung-Shan Institute of Science and Technology (NCSIST) and the Coast Guard Administration attending.
Thunder Tiger Corporation supplied the vessels – a SeaShark 600 and a larger SeaShark 800 – while Shield AI supplied the autonomy stack, which planned, searched and coordinated the escort using maritime radar, imagery and automatic identification system (AIS) data.
Hivemind divided the search box into two sectors and assigned one vessel to each, and when the first USV made the detection, real-time data sharing allowed the second to close on an optimized intercept route.
The vessel driven off was crewed, and was expelled without human intervention across a test area of roughly two square kilometres.
Shield AI also described the event as its first integration of maritime radar and AIS into Hivemind, which has now flown or sailed on more than thirty platforms, including the Kratos MQM-178 Firejet.
What stands out is the pace. The two firms signed on 13 May 2026 and committed to a live demonstration in summer 2026, moving through simulation, then hardware-in-the-loop, then live vehicle.
Integration of the software onto the vessels was completed in roughly six weeks.
That timeline follows Shield AI’s February 2026 contract with NCSIST, covering the Hivemind software development kit for indigenous Taiwanese platforms alongside turnkey Hivemind for NCSIST multi-role unmanned systems.
Thunder Tiger’s own hardware line predates the partnership. The SeaShark 800 was unveiled in April 2025 at 8.1 m in length and 2.7 tonnes full displacement, with a stated 600 km range, speeds above 50 knots and a low-radar-cross-section hull.
The 6 m SeaShark 600 Stealth followed in September 2025 with an aluminium-alloy hull roughly half the weight of fibre-reinforced plastic and a payload of 300 to 600 kg.
The company took a controlling stake in Morningstar Boats in September 2024 and now builds unmanned surface vehicles at Luzhu Industrial Park in Taoyuan at about one hull per week, with a further plant at Chiayi fitting out in the second half of 2026.
Autonomy is the newer variable. In November 2025 Thunder Tiger ran six SeaShark 680 USVs from a single control station at the same Dapeng Bay site, holding line, lightning and single-file formations – more hulls than the July event, but with a human directing the pattern rather than software selecting it.
That progression matters because the barrier to fielding USVs in numbers has shifted from hulls to software, a pattern visible in the attack USV designs that struck Bandar Abbas.
Other navies have reached the same conclusion, with Pakistan moving to develop its own autonomy stacks alongside the airframes and hulls that carry them.
The commercial target is a procurement line that has not yet cleared Taiwan’s legislature. The Executive Yuan’s June 2026 special budget of NT$210 billion, or roughly US$6.6 billion, covers 208,200 attack drones, 1,446 reconnaissance drones and 1,320 unmanned surface vessels.
The USV portion traces to NCSIST’s Kuai Chi program, worth about NT$18 billion over five years and divided between the Navy’s Littoral Combat Command, the Marine Corps and Army Special Operations Command, with Lungteh, Jong Shyn and CSBC also competing.
Taiwan is not alone in ordering expendable hulls at that scale, with the Turkish Navy separately moving to procure 100 expendable USVs from Aselsan, STM and Havelsan for swarm strike operations.
No SeaShark order has been publicly confirmed against the Taiwanese line, which remains pending in the Legislative Yuan.
Shield AI’s director of government relations for Asia, Maximilian Chen, told reporters at the demonstration that five USVs could cover up to 8.3 times their combined search area if Hivemind were used.
