The ocean has always been a place where machines disappear. A submarine that descends to operational depth leaves behind the electromagnetic spectrum that modern warfare depends on — GPS signals, radio frequencies, satellite links, real-time command networks. It enters a domain governed by pressure, thermoclines, salinity gradients, and the physics of sound. For decades, this isolation was managed by human crews who exercised judgment, adapted to conditions, and radioed home when they surfaced.
Now the crews are being removed. What remains is an AI system navigating a three-dimensional battlespace with no connection to the surface, no GPS waypoint corrections, and no ability to call for instructions. It must identify targets, avoid threats, manage its energy budget, and in some configurations, execute lethal missions entirely on its own.
The autonomous underwater vehicle has matured from a curiosity into a weapons platform. The programs now operating or in advanced development across the United States, China, Australia, and Russia represent a fundamental restructuring of undersea power — one that will reshape naval doctrine, arms control frameworks, and strategic stability in every ocean simultaneously.
Boeing Orca XLUUV: The Pentagon's 51-Foot Robot Submarine
The Extra-Large Unmanned Undersea Vehicle, or XLUUV, is the Pentagon's most advanced autonomous submarine program. Boeing was awarded a $43 million contract for the Orca in 2019, followed by a $274 million contract for five vehicles delivered through 2022. The program has since expanded, with the Navy committing to a larger fleet as testing progressed through 2024 and 2025.
The Orca is 51 feet long — roughly the length of a transit bus — and displaces approximately 50 metric tons when fully loaded. That scale places it in an entirely different category from the small-diameter torpedo-shaped UUVs that navies have operated for years. The Orca can dive to depths exceeding 11,000 feet, operate for months without a crew, and travel thousands of nautical miles on a single mission before returning to a support vessel or pier-side facility for resupply.
The vehicle's payload section is modular by design. The baseline configuration carries a sensing suite optimized for intelligence, surveillance, and reconnaissance: passive sonar arrays, synthetic aperture sonar for seafloor mapping, and signals intelligence sensors that can characterize electromagnetic activity in the surrounding water column. In the mine warfare configuration — which the Navy has prioritized — the Orca carries and emplaces Hammerhead mines, a torpedo-capable weapon that sits on the seafloor and engages surface vessels or submarines passing overhead. A single Orca could seed an entire strait or approach channel with networked mines in a single mission, then return to reload.
The AI system driving the Orca's autonomous behavior is built around Boeing's Autonomy Kit, a modular software architecture that handles mission planning, navigation, threat avoidance, and payload management. The system uses a combination of inertial navigation, terrain-relative navigation against pre-loaded seafloor maps, and acoustic self-localization against known underwater features. It does not need GPS. It uses the seafloor itself as a reference grid.
Anduril Dive-LD: The Software-Defined Submarine
Where Boeing's Orca was built through traditional defense contracting channels, Anduril's Dive-LD represents the new-generation approach to autonomous submarine development: rapid iteration, modular payload architecture, and software-defined mission systems that can be updated over the air between deployments.
The Dive-LD, first disclosed publicly in 2022 following Anduril's acquisition of Dive Technologies, is a mid-sized autonomous underwater vehicle designed for maximum flexibility. Its torpedo-like hull can accommodate a range of payload configurations, from ISR sensor packages to electronic warfare systems to weapons carriage. The vehicle's defining characteristic is its Lattice software platform — the same AI-driven command-and-control system that runs Anduril's Ghost surface vessels and Roadrunner aerial systems — adapted for the underwater environment.
Lattice's integration means the Dive-LD is not merely an autonomous vehicle; it is a node in a broader AI network. In exercises conducted through 2024, Anduril demonstrated the Dive-LD operating in concert with Ghost surface vessels and Altius-700M aerial drones, sharing acoustic contact data, coordinating positioning, and presenting a fused picture of the battlespace to an offshore command center. The undersea vehicle feeds its acoustic detections to the surface node, which elevates them via satellite to the command network. The human operator sees the compiled picture without ever needing to query the individual vehicles.
The Dive-LD's payload modularity has attracted significant Navy interest. The vehicle can be reconfigured between missions in hours rather than days, allowing a single pool of vehicles to serve different mission sets as operational requirements change. Its relatively small logistics footprint — it can be launched from a standard pier, a support vessel, or a modified submarine launch tube — makes it applicable across theater contexts, from the Pacific island chains to the Arctic littoral.
Manta Ray: The Long-Endurance Seabed Platform
The Manta Ray program, developed by Northrop Grumman under a DARPA contract awarded in 2020, occupies a different operational concept from the Orca and Dive-LD. Rather than executing discreet missions and returning, the Manta Ray is designed for persistent forward presence — the ability to operate in a contested area for months or years, drawing power from ocean currents and thermal gradients through energy harvesting systems, and activating only when triggered by a pre-programmed condition or a remote command.
The vehicle's shape, which resembles the manta ray after which it is named with its wide, flat body and articulated pectoral fins, is optimized for low acoustic signature and efficient hydrodynamic performance at slow speeds. Northrop Grumman successfully completed in-water testing of the full-scale Manta Ray prototype in 2024, validating the vehicle's buoyancy control, energy harvesting, and low-power navigation systems in real-ocean conditions off the coast of Southern California.
The Manta Ray's operational concept upends traditional forward deployment logistics. Rather than transiting a submarine or surface ship to a contested area, the Navy could pre-position Manta Ray vehicles on the seafloor near strategic chokepoints months or years before a conflict begins. When hostilities commence, those vehicles could be activated remotely, execute their missions, and either return for recovery or continue operating until their energy reserves are exhausted. The concept is sometimes called "seabed warfare" — treating the ocean bottom as a maneuver space rather than merely a surface over which other platforms operate.
The Manta Ray's energy harvesting capability draws on ocean thermal energy conversion, wave energy, and thermoelectric systems that exploit temperature differentials in the water column. In favorable deep-water environments, these systems could theoretically extend vehicle endurance indefinitely, limited only by component wear rather than fuel. This represents a fundamental break from every previous underwater military platform, which was constrained by the chemistry of its battery pack or the capacity of its fuel tanks.
Ghost Shark: AUKUS Pillar II and the Australian Strategic Bet
The Ghost Shark program, developed by Anduril Australia in partnership with the Australian Defence Force, represents the most significant non-American autonomous submarine initiative in the Western alliance. Announced in February 2024 as an AUKUS Pillar II project — the technology-sharing track of the Australia-UK-US partnership — Ghost Shark is designed to give Australia an undersea autonomous capability that functions as a force multiplier for the relatively small Royal Australian Navy.
Australia's strategic geography makes autonomous underwater vehicles particularly relevant. The continent is surrounded by vast maritime approaches. The distances from Australian ports to the most likely conflict zones in the South China Sea and along the first island chain exceed what crewed submarine patrols can cover with the frequency a deterrence posture requires. A fleet of Ghost Shark autonomous vehicles could maintain persistent presence across those approaches, feeding intelligence to Australian and allied command networks and presenting a threat to adversary surface vessels and submarines that would otherwise operate with impunity in the region.
Ghost Shark is a large-displacement UUV designed to carry a substantial payload. Its hull is derived from Anduril's broader autonomous platform architecture, with the Lattice operating system enabling integration with Australian naval networks and allied command structures. The vehicle can operate independently for extended periods or as part of a coordinated autonomous swarm. In the swarm configuration, multiple Ghost Shark vehicles coordinate their sensor coverage, share contact data, and present coordinated cueing to human decision-makers without requiring individual teleoperation.
The AUKUS context gives Ghost Shark a strategic dimension beyond its tactical capabilities. Australia's investment in autonomous submarine technology is explicitly designed to offset the delay in delivering nuclear-powered submarines — which under the AUKUS agreement are not expected to reach Royal Australian Navy service until the mid-2030s at the earliest. Ghost Shark is the interim capability that allows Australia to develop the doctrine, the operators, the maintenance infrastructure, and the command integration that will be needed when nuclear submarines eventually arrive.
China's Underwater AI Fleet: Sea Wing, Haiyi, and the South China Sea
While Western programs have attracted the most public attention, China's autonomous underwater vehicle program has quietly become the most numerically significant in the world. Chinese researchers and defense institutions have deployed hundreds of underwater gliders — autonomous vehicles that navigate using changes in buoyancy rather than propellers — across the South China Sea, Western Pacific, and Indian Ocean. The scale of this deployment has no equivalent in any other navy.
The two primary Chinese systems are the Sea Wing (Haiyi) glider, developed by the Chinese Academy of Sciences' Shenyang Institute of Automation, and the HN-1 underwater glider developed by Tianjin University. Both systems are nominally civilian oceanographic research platforms, and both have been deployed in vast numbers that exceed any plausible civilian scientific requirement. Western intelligence assessments uniformly conclude that the data being collected by these vehicles includes military-relevant information: water temperature and salinity profiles that determine sonar propagation characteristics, seafloor bathymetry in areas of strategic significance, and tracking data on surface vessel movements.
In 2016, a Chinese Navy vessel recovered an American UUV that had been deployed by the USNS Bowditch research ship in the South China Sea. The incident was diplomatically embarrassing and reinforced Western concerns about the legal and operational ambiguity of autonomous underwater systems in contested waters. China's response, which included the eventual return of the vehicle, also demonstrated that Chinese naval forces were actively monitoring and capable of recovering foreign underwater platforms operating in the region.
China has also invested heavily in larger combat-capable autonomous undersea systems. The HSU001 autonomous underwater vehicle, unveiled at the 2019 National Day parade, is a blunt-nosed, propeller-driven UUV roughly 3 meters in length that carries an optical sensor suite and is designed for undersea reconnaissance and target acquisition. Its existence confirms that China has moved beyond civilian oceanographic platforms to dedicated military AUV development. Chinese military publications have described concepts for autonomous underwater vehicles carrying torpedoes, mines, and electronic warfare payloads — systems that would dramatically expand the reach and persistence of People's Liberation Army Navy (PLAN) undersea operations without requiring the construction of additional crewed submarines.
"China has deployed more autonomous underwater systems in the South China Sea than the rest of the world's navies combined. The data they are collecting is shaping their understanding of the acoustic environment in every significant chokepoint in the Western Pacific."
-- Former U.S. Pacific Fleet Intelligence Officer, 2025
Russia's Poseidon: The Nuclear Autonomous Torpedo
No program in the autonomous undersea domain is as strategically consequential, or as deliberately destabilizing, as Russia's Poseidon torpedo, known in official Russian military documentation as the Status-6 system and in NATO parlance as the Kanyon. Poseidon is not a vehicle that carries conventional weapons. It is a nuclear-armed autonomous torpedo — approximately 65 feet long, driven by a nuclear propulsion system, and designed to carry a thermonuclear warhead with a yield estimated by U.S. intelligence at up to 100 megatons.
The weapon was first publicly disclosed by Russian President Vladimir Putin in a March 2018 address to the Federal Assembly, in which he used animated video to illustrate the weapon evading American missile defenses and striking coastal cities. The strategic concept is straightforward and horrifying: Poseidon is designed to be launched from a modified submarine — the Belgorod-class SSBN, which entered service in 2022, is the primary carrier — and then navigate autonomously across ocean basins to detonate near enemy coastlines, generating a massive radioactive tsunami that would contaminate coastal cities for decades.
Poseidon's autonomous navigation is enabled by the same physics that makes it immune to existing countermeasures. It travels at depths that defeat existing torpedo defenses, at speeds that existing fast-torpedo countermeasures cannot intercept, and along routes that are determined by its on-board AI system without real-time guidance from its launch platform. Once deployed, the weapon cannot be recalled through conventional means. It operates entirely autonomously from launch to detonation.
The arms control implications of Poseidon are profound. Existing nuclear arms control treaties — including New START, which lapsed in 2026 — counted nuclear warheads on delivery vehicles such as ICBMs and submarine-launched ballistic missiles. Poseidon occupies an ambiguous category: it is an autonomous undersea vehicle that carries a nuclear warhead, but it is not formally classified as a ballistic missile or a nuclear-armed cruise missile. Whether it falls under any existing arms control framework, and how any future framework could verify and limit Poseidon deployments, are questions that strategic arms control experts have not resolved.
Poseidon represents the convergence of two arms control nightmares: autonomous weapons systems and nuclear weapons. The combination produces a strategic system that, once launched, operates outside any command-and-control structure. There is no nuclear hotline that can stop it. There is no recall code. The only deterrent is the threat of assured retaliation before launch — a calculus that depends on detecting the submarine carrier before it deploys the weapon.
The Unique AI Challenges of the Underwater Domain
Building an autonomous system that operates in the air or on land is difficult. Building one that operates underwater is categorically harder. The ocean imposes constraints on autonomous systems that have no equivalent in any other operational domain, and understanding those constraints is essential to understanding both the current limitations of autonomous underwater vehicles and the trajectory of their development.
No GPS, No Radio: The Navigation Problem
GPS signals do not penetrate seawater. Neither do the radio frequencies that allow aerial and ground drones to receive real-time commands, upload updated intelligence, or communicate with other platforms. An autonomous underwater vehicle, once it submerges, is effectively cut off from the electromagnetic infrastructure that all other modern autonomous systems depend on.
This forces AUVs to rely on inertial navigation systems — gyroscopes and accelerometers that track the vehicle's position by integrating its acceleration over time. Inertial navigation accumulates error: small measurement inaccuracies compound over time, so a vehicle that has been underwater for several days may have a position error measured in hundreds of meters or more. Correcting this error requires the vehicle to either surface and acquire a GPS fix (which exposes it to detection), use acoustic transponder networks that must be pre-positioned in the operational area, or use terrain-relative navigation against pre-loaded seafloor maps.
Terrain-relative navigation — matching acoustic bathymetric measurements against a stored seafloor chart — is the most operationally promising solution for long-range missions in areas where the seafloor topography is well-mapped. The challenge is that large areas of the world's oceans, particularly in the Western Pacific and Arctic, are poorly charted. The same Chinese underwater glider deployments that raise military concerns are also generating the detailed seafloor maps that could enable future autonomous underwater operations in those regions.
Acoustic-Only Sensing: The Sonar Challenge
In the air, autonomous vehicles can use optical cameras, radar, lidar, and infrared sensors to perceive their environment. Underwater, the only sensor that works at meaningful ranges is sonar — active or passive acoustic systems that either emit sound pulses and listen for returns, or listen passively for sounds generated by other vessels and marine life. Sonar is powerful but also constrained.
Active sonar reveals the position of the emitting vehicle as clearly as it reveals the position of its targets. An autonomous submarine that uses active sonar to navigate or acquire targets announces its presence to any adversary system in the area. Passive sonar avoids this problem but is limited in range and dependent on the acoustic environment — the temperature gradients, salinity layers, and bottom conditions that determine how sound propagates in a given area.
Machine learning has substantially improved the performance of both active and passive sonar systems. Neural networks trained on large libraries of acoustic signatures can identify specific vessel classes, estimate their bearing and speed, and track them through acoustic clutter with a reliability that exceeds human operators. But the training data itself is a limiting factor: acoustic signatures vary by ocean region, season, and vessel loading, and a model trained on Pacific acoustic environments may perform poorly in the Arctic or Indian Ocean without retraining on locally collected data.
Pressure, Currents, and the Physics of Depth
Water pressure increases by approximately one atmosphere for every ten meters of depth. At 1,000 meters — well within the operational range of modern AUVs — the pressure exceeds 100 atmospheres, or roughly 1,500 pounds per square inch. Electronics must be housed in pressure-rated enclosures. Seals must be maintained against continuous hydrostatic load. Actuators must work reliably under conditions that degrade mechanical systems rapidly.
Ocean currents add further complexity. An AUV navigating against a 2-knot current will arrive at its intended waypoint with a significantly different battery state than one operating in calm water. Currents vary in strength and direction unpredictably, particularly in shelf regions, straits, and areas near major river outflows. An autonomous system's power management AI must continuously balance mission completion against energy reserves, and it must do so without the ability to request additional power or call for rescue if something goes wrong.
Autonomous Mine Warfare: The Sleeper Threat
Of all the missions that autonomous underwater vehicles will perform, the one with the most immediate tactical significance is mine warfare. Naval mines have been among the most cost-effective weapons in modern naval history — a $10,000 mine can sink a $2 billion destroyer — and the combination of autonomous delivery systems with intelligent mine technology is producing a capability that could close major maritime chokepoints at a fraction of the cost of any other denial system.
The United States has developed the Hammerhead system specifically for this concept: a torpedo-armed seafloor mine that can be emplaced by autonomous vehicles such as the Orca, remain dormant for extended periods, and engage surface vessels or submarines passing within its detection envelope using its torpedo payload. The system is essentially a distributed autonomous weapons platform: once emplaced, each mine makes its own targeting decisions based on its acoustic sensors without any connection to a command network.
The implications for Pacific contingency planning are significant. The U.S. Navy has publicly stated that one of its primary objectives in a Taiwan contingency would be to deny the People's Liberation Army Navy freedom of movement through the Luzon Strait, the Miyako Strait, and the Bashi Channel — the three primary exits from Chinese home waters into the Western Pacific. Autonomous mine-laying vehicles could seed those straits with Hammerhead mines in the days or hours before a conflict, creating denial barriers that would be extraordinarily expensive and time-consuming for Chinese minesweeping forces to clear.
China is pursuing parallel capabilities. Chinese military publications have described autonomous mine-laying concepts using underwater gliders and larger AUVs. The South China Sea, where China has constructed and fortified artificial islands, is a natural environment for Chinese autonomous mine warfare — the shallow water depths, well-mapped bathymetry, and Chinese operational familiarity with the area all favor the defender's use of autonomous mining systems.
The Anti-Submarine Warfare Revolution
For most of the Cold War, anti-submarine warfare was the Navy's most demanding and resource-intensive operational challenge. Detecting a modern nuclear submarine in the open ocean required a layered network of fixed hydrophone arrays, maritime patrol aircraft, helicopter-equipped surface vessels, and hunter-killer submarines — a system that consumed enormous resources and still left significant detection gaps.
Autonomous systems are beginning to transform this calculus. The Navy's Sea Hunter program — an autonomous surface vessel designed specifically to track diesel-electric submarines by exploiting their intermittent acoustic signatures over extended periods — demonstrated in trials that a single unmanned surface vessel could maintain contact with a submarine for days, covering thousands of square miles of ocean at a cost per operating hour that is a small fraction of what a manned patrol aircraft or destroyer requires. The Orca and similar AUVs extend this capability underwater, enabling passive sonar surveillance in areas where surface vessels or fixed arrays cannot operate without detection.
The combination of AUV-based undersea surveillance, autonomous surface vessel tracking, and AI-driven acoustic analysis is creating what Navy analysts describe as a persistent undersea awareness architecture — a distributed sensor network that provides continuous tracking of adversary submarine activity across vast ocean areas. China has recognized this emerging capability and has responded by investing heavily in quieting technology for its submarine fleet, developing decoys and acoustic masking systems, and pursuing its own autonomous ASW platforms.
From SOSUS to AI: The Intelligence Legacy
The conceptual foundation for AI-driven undersea surveillance dates to the Cold War's most important secret program: the Sound Surveillance System, or SOSUS. Built beginning in the 1950s, SOSUS was a network of hydrophone arrays cabled to shore stations on the continental margins of the Atlantic and Pacific, designed to detect the acoustic signatures of Soviet submarines transiting toward American waters. At its peak, SOSUS could track Soviet nuclear submarines across entire ocean basins, providing warning of potential strikes that was critical to American nuclear deterrence posture.
The physics of SOSUS detection exploited a natural underwater acoustic channel called the SOFAR (Sound Fixing and Ranging) layer — a depth band at which sound waves travel with minimum attenuation, allowing them to propagate thousands of miles. American hydrophone arrays listening in the SOFAR channel could detect submarine propeller noise from distances that seemed impossibly large. The Soviets knew SOSUS existed but could not consistently evade it, which constrained their operational doctrine and forced investments in quieter submarine designs.
SOSUS's limitations were its fixed geography and its dependence on manual analysis of acoustic data by trained operators. The system covered the Atlantic and specific Pacific transit routes well; it had significant gaps in the Western Pacific, Arctic, and Indian Ocean. And the acoustic data it collected required skilled human analysts who could distinguish submarine signatures from biological noise, shipping traffic, and geological events — a process that was slow and labor-intensive even at its most effective.
Modern AI-driven acoustic analysis eliminates the human analysis bottleneck. Neural networks trained on decades of acoustic data — including historical SOSUS recordings — can classify contacts, track them through complex acoustic environments, and alert human operators to significant detections in seconds rather than hours. When combined with the distributed sensor networks that autonomous underwater vehicles can create, AI-driven acoustic analysis transforms undersea surveillance from a fixed-geography asset into a flexible, forward-deployable capability.
Doctrine, Law, and the Accountability Gap
The rapid proliferation of autonomous underwater vehicles has outpaced the development of any coherent legal or doctrinal framework governing their use. This gap creates significant risks in the near term and potentially catastrophic consequences in the longer term.
International law of the sea does not clearly address autonomous underwater vehicles. The UN Convention on the Law of the Sea establishes rights of innocent passage, freedom of navigation, and the legal status of military vessels, but was written before autonomous systems existed. There is no international consensus on whether an autonomous underwater vehicle conducting surveillance in another nation's exclusive economic zone violates international law. China has claimed that American UUV operations in the South China Sea are illegal; the United States has maintained they are consistent with freedom of navigation rights. There is no tribunal or enforcement mechanism to resolve this dispute.
The rules of engagement for autonomous underwater combat systems are equally unresolved. A Hammerhead mine, once emplaced, cannot distinguish between an adversary warship and a civilian vessel. An autonomous torpedo-armed AUV operating in a contested area must determine, without human input, whether a detected contact meets the criteria for engagement. The existing framework of international humanitarian law — requiring discrimination between combatants and civilians, proportionality, and precaution — was designed for humans making decisions. Applying it to algorithms has not been adequately addressed by any major military legal system.
The United States has taken the position that a human must be "in the loop" or "on the loop" for lethal decisions made by autonomous systems — a principle established in DoD Directive 3000.09 and maintained, with some erosion, through the Hegseth directives of 2026. But autonomous mine warfare already challenges this principle: a mine that activates and fires autonomously when a vessel triggers its sensor is not subject to real-time human authorization, and yet mines have been a legal weapon of war for over a century.
The Next Decade: What Comes After the Orca
The programs described in this report — Orca, Dive-LD, Manta Ray, Ghost Shark, Poseidon — represent the first generation of purpose-built military autonomous underwater vehicles. The second generation, which will enter development over the next five years, will be substantially more capable across every dimension.
Advances in battery energy density, drawing on the same lithium-sulfur and solid-state battery research that is transforming electric vehicles, will dramatically extend AUV endurance. Improvements in acoustic signal processing, driven by transformer-model architectures applied to sonar data, will improve detection range and classification accuracy. Advances in distributed AI, enabling vehicle swarms to coordinate without a centralized node, will make large-scale autonomous undersea operations feasible without the communication infrastructure that current swarm concepts depend on.
The geopolitical trajectory is equally clear. China will continue expanding its autonomous underwater presence in the South China Sea and Western Pacific, using the oceanographic data it is collecting to optimize AUV operations in those waters. Russia will continue developing Poseidon and related autonomous nuclear systems as a strategic hedge against American conventional military superiority. AUKUS will generate additional collaborative autonomous undersea programs, deepening interoperability between the three partners. And smaller navies — from Norway to Iran to India — will increasingly field commercially available AUV platforms adapted for military use, democratizing a capability that was once the exclusive province of great powers.
The ocean floor is becoming a battlespace. The vehicles that are learning to fight there carry no crews to rescue, generate no diplomatic crises when lost, and can be deployed in numbers that crewed submarine fleets cannot approach. The strategic implications will take decades to fully understand. But the silent war below has already begun.