SPECIAL REPORT-In U.S.-China AI contest, the race is on to deploy killer robots


By David Lague

SYDNEY, Sept 8 (Reuters) - To meet the challenge of a rising China, the Australian Navy is taking two very different deep dives into advanced submarine technology.

One is pricey and slow: For a new force of up to 13 nuclear-powered attack submarines, the Australian taxpayer will fork out an average of more than AUD$28 billion ($18 billion) apiece. And the last of the subs won’t arrive until well past the middle of the century.

The other is cheap and fast: launching three unmanned subs, powered by artificial intelligence, called Ghost Sharks. The navy will spend just over AUD$23 million each for them – less than a tenth of 1% of the cost of each nuclear sub Australia will get. And the Ghost Sharks will be delivered by mid-2025.

The two vessels differ starkly in complexity, capability and dimension. The uncrewed Ghost Shark is the size of a school bus, while the first of Australia’s nuclear subs will be about the length of a football field with a crew of 132. But the vast gulf in their cost and delivery speed reveal how automation powered by artificial intelligence is poised to revolutionize weapons, warfare and military power – and shape the escalating rivalry between China and the United States. Australia, one of America’s closest allies, could have dozens of lethal autonomous robots patrolling the ocean depths years before its first nuclear submarine goes on patrol.

Without the need to cocoon a crew, the design, manufacture and performance of submarines is radically transformed, says Shane Arnott. He is the senior vice-president of engineering at U.S. defense contractor Anduril, whose Australian subsidiary is building the Ghost Shark subs for the Australian Navy.

“A huge amount of the expense and systems go into supporting the humans,” Arnott said in an interview in the company’s Sydney office.

Take away the people, and submarines become much easier and cheaper to build. For starters, Ghost Shark has no pressure hull – the typically tubular, high-strength steel vessel that protects a submarine's crew and sensitive components from the immense force that water exerts at depth. Water flows freely through the Ghost Shark structure. That means Anduril can build lots of them, and fast.

Rapid production is the company’s plan. Arnott declined to say, though, how many Ghost Sharks Anduril intends to manufacture if it wins further Australian orders. But it is designing a factory to build “at scale,” he said. Anduril is also aiming to build this type of sub for the United States and its allies, including Britain, Japan, Singapore, South Korea and customers in Europe, the company told Reuters.

A need for speed is driving the project. Arnott points to an Australian government strategic assessment, the Defense Strategic Review, published in April, which found the country was entering a perilous period where “China's military build-up is now the largest and most ambitious of any country since the end of the Second World War.” A crisis could emerge with little or no warning, the review said.

“We can’t wait five to 10 years, or decades, to get stuff,” said Arnott. “The timeline is running out.”

This report is based on interviews with more than 20 former American and Australian military officers and security officials, reviews of AI research papers and Chinese military publications, as well as information from defense equipment exhibitions.


An intensifying military-technology arms race is heightening the sense of urgency. On one side are the United States and its allies, who want to preserve a world order long shaped by America’s economic and military dominance. On the other is China, which rankles at U.S. ascendancy in the region and is challenging America’s military dominance in the Asia-Pacific. Ukraine’s innovative use of technologies to resist Russia’s invasion is heating up this competition.

In this high-tech contest, seizing the upper hand across fields including AI and autonomous weapons, like Ghost Shark, could determine who comes out on top.

“Winning the software battle in this strategic competition is vital,” said Mick Ryan, a recently retired Australian army major general who studies the role of technology on warfare and has visited Ukraine during the war. “It governs everything from weather prediction, climate change models, and testing new-era nuclear weapons to developing exotic new weapons and materials that can provide a leap-ahead capability on the battlefield and beyond.”

If China wins out, it will be well placed to reshape the global political and economic order, by force if necessary, according to technology and military experts.

Most Americans alive today have only known a world in which the United States was the single true military superpower, according to a May report, Offset-X, from the Special Competitive Studies Project, a non-partisan U.S. panel of experts headed by former Google Chairman Eric Schmidt. The report outlines a strategy for America to gain and maintain dominance over China in military technology.

If America fails to act, it “could see a shift in the balance of power globally, and a direct threat to the peace and stability that the United States has underwritten for nearly 80 years in the Indo-Pacific,” the report said. “This is not about the anxiety of no longer being the dominant power in the world; it is about the risks of living in a world in which the Chinese Communist Party becomes the dominant power.”

The stakes are also high for Beijing. If the U.S. alliance prevails, it will make it far harder for the People’s Liberation Army, or PLA, as the Chinese military is known, to seize democratically governed Taiwan, control the shipping lanes of East Asia and dominate its neighbors. Beijing sees Taiwan is an inalienable part of China and hasn’t ruled out the use of force to subdue it.

The Department of Defense had no comment “on this particular report,” a Pentagon spokesperson said in response to questions. But the department’s leadership, the spokesperson added, has been “very clear” regarding China as “our pacing challenge.” Regarding a possible attack on Taiwan, the spokesperson said, Defense Secretary Lloyd Austin and other senior leaders “have been very clear that we do not believe an invasion is imminent or inevitable, because deterrence today is real and strong.”

China’s defense ministry and foreign ministry didn’t respond to questions for this article.

A spokesperson for Australia’s Department of Defence said it was a priority to “translate disruptive new technologies into Australian Defence Force capability.” The department is investigating, among other things, “autonomous undersea warfare capabilities to complement its crewed submarines and surface fleet, and enhance their lethality and survivability,” the spokesperson said.


Some leading military strategists say AI will herald a turning point in military power as dramatic as the introduction of nuclear weapons. Others warn of profound dangers if AI-driven robots begin making lethal decisions independently, and have called for a pause in AI research until agreement is reached on regulation related to the military application of AI.

Despite such misgivings, both sides are scrambling to field uncrewed machines that will exploit AI to operate autonomously: subs, warships, fighter jets, swarming aerial drones and ground combat vehicles. These programs amount to the development of killer robots to fight in tandem with human decision makers.

Such robots – some designed to operate in teams with conventional ships, aircraft and ground troops – already have the potential to deliver sharp increases in firepower and change how battles are fought, according to military analysts.

Some, like Ghost Shark, are able to perform maneuvers no conventional military vehicle could survive – like diving thousands of meters below the ocean surface.

Perhaps even more revolutionary than autonomous weapons is the potential for AI systems to inform military commanders and help them decide how to fight – by absorbing and analyzing the vast quantities of data gathered from satellites, radars, sonar networks, signals intelligence and online traffic. Technologists say this information has grown so voluminous it is impossible for human analysts to digest. AI systems trained to crunch this data could deliver commanders with better and faster understanding of a battlefield and provide a range of options for military operations.

Conflict may also be on the verge of turning very personal. The capacity of AI systems to analyze surveillance imagery, medical records, social media behavior and even online shopping habits will allow for what technologists call “micro-targeting” – attacks with drones or precision weapons on key combatants or commanders, even if they are nowhere near the front lines. Kiev’s successful targeting of senior Russian military leaders in the Ukraine conflict is an early example.

AI could also be used to target non-combatants. Scientists have warned that swarms of small, lethal drones could target big groups of people, such as the entire population of military-aged males from a certain town, region or ethnic group.

“They could wipe out, say, all males between 12 and 60 in a city,” said computer scientist Stuart Russell in a BBC lecture on the role of AI in warfare broadcast in late 2021. “Unlike nuclear weapons, they leave no radioactive crater, and they keep all the valuable physical assets intact,” added Russell, a professor of computer science at the University of California, Berkeley.

The United States and China have both tested swarms of AI-powered drones. Last year, the U.S. military released footage of troops training with drone swarms. Another video shows personnel at Fort Campbell, Tennessee, testing swarms of drones in late 2021. The footage shows a man wearing video game-like goggles during the experiment.

For the U.S. alliance, swarms of cheap drones could offset China’s numerical advantage in missiles, warships and strike aircraft. This could become critical if the United States intervened against an assault by Beijing on Taiwan.

America will field “multiple thousands” of autonomous, unmanned systems within the next two years in a bid to offset China's advantage in numbers of weapons and people, the U.S. Deputy Secretary of Defense, Kathleen Hicks, said in an August 28 speech. “We’ll counter the PLA’s mass with mass of our own, but ours will be harder to plan for, harder to hit, harder to beat,” she said.

Even drones with limited AI capability can have an impact. Miniature, remote-controlled surveillance drones with some autonomy are already in service. One example is the pocket-sized Black Hornet 3 now being deployed by multiple militaries.

This drone can fit in the palm of a hand and is hard to detect, according to the website of Teledyne FLIR, the company that makes them. It is reminiscent of the movie “Eye in the Sky,” in which a bug-like drone is used against militants in Kenya. Weighing less than 33 grams, or a bit more than an ounce, it can fly almost silently for 25 minutes, sending back video and high-definition still images to its operator. It gives soldiers in the field a real-time understanding of what is happening around them, according to the company.


The AI military sector is dominated by software, an industry where change comes fast.

Anduril, maker of the AI-powered Ghost Shark, is trying to capitalize on the desire of the U.S. alliance to quickly team humans with intelligent machines. The company, which shares its name with a fictional sword in Tolkien’s “Lord of the Rings” saga, was founded in 2017 by Palmer Luckey, designer of the Oculus virtual reality headset, now owned by Facebook. Luckey sold Oculus VR to the social media giant for $2.3 billion in 2014.

Arnott, the Anduril engineer working on Ghost Shark, said the company is also supplying equipment to Ukraine. The Russians rapidly adapted to this gear deployed in battle, so Anduril has been pushing out regular updates to maintain an advantage.

“Something happens,” he said. “We get punched in the face. The customer gets hit with something, and we are able to take that, turn it around and push out a new feature.”

Arnott didn’t provide details of the equipment, but Anduril referred Reuters to a February announcement from the Biden administration that included the company’s ALTIUS 600 munition drone in a package of military aid to Ukraine. This drone can be deployed for intelligence, surveillance and reconnaissance. It can also be used as a kamikaze drone, armed with an explosive warhead that can fly into enemy targets.

Ukraine has already reportedly used drone surface craft packed with explosives to attack Russian shipping. Military commentators have suggested that Taiwan could use similar tactics to resist a Chinese invasion, launching big numbers of these vessels into the path of the fleet heading for its beaches.

Asked by Reuters about Taiwan’s drone program, the office of President Tsai Ing-wen said in June that the island had drawn “great inspiration” from Ukraine’s use of drones in its war with Russia.

China, the United States and U.S. allies have programs to build fleets of stealthy drone fighters that will fly in formation with crewed aircraft. The drones could peel off to attack targets, jam communications or fly ahead so their radars and other sensors could provide early warning or find targets. These robots could instantly share information with each other and human operators, according to military technology specialists.

America is planning to build a 1,000-strong fleet of these fighter drones, U.S. Air Force Secretary Frank Kendall told a warfare conference in Colorado in March. At the Zhuhai air show in November, China unveiled a jet fighter-like drone, the FH-97A, which will operate with a high degree of autonomy alongside manned combat aircraft, providing intelligence and added firepower, according to reports in the Chinese state-controlled media. China, the United States and Japan are also building large, uncrewed submarines similar to Australia’s Ghost Shark.

One overwhelming advantage of these autonomous weapons: Commanders can deploy them in big numbers without risking the lives of human crews. In some respects, performance improves, too.

Jet-powered robot fighters, for instance, could perform maneuvers the human body wouldn’t tolerate. This would include tight turns involving extreme G-forces, which can cause pilots to pass out. Aerial drones can also do away with the pressurized cockpits, oxygen supplies and ejector seats required to support a human pilot.

And robots don’t get tired. As long as they have power or fuel, they can carry on their missions indefinitely.

Because many robots are relatively cheap – a few million dollars for an advanced fighter drone, versus tens of millions for a piloted fighter jet – losses could be more readily absorbed. For commanders, that means more risk might become acceptable. A robot scout vehicle could approach an enemy ground position to send back high-definition images of defenses and obstacles, even if it is subsequently destroyed, according to Western military experts.


So far, it is difficult to say who is winning the battle to master AI-powered weapons. China’s huge and sophisticated manufacturing sector gives it advantages in mass production. America remains home to most of the world’s dominant and most innovative technology and software companies. But tight secrecy surrounds the projects on both sides.

Beijing does not publish any detailed breakdown of its increasing defense spending, including outlays on AI. Still, the available disclosures of spending on AI military research do show that outlays on AI and machine learning grew sharply in the decade from 2010.

In 2011, the Chinese government spent about $3.1 million on unclassified AI research at Chinese universities and $8.5 million on machine learning, according to Datenna, a Netherlands-based private research company specializing in open source intelligence on China's industrial and technology sectors. By 2019, AI spending was about $86 million and outlays on machine learning were about $55 million, Datenna said.

“The biggest challenge is we don’t really know how good the Chinese are, particularly when it comes to the military applications of AI,” said Martijn Rasser, a former analyst with the U.S. Central Intelligence Agency and now managing director of Datenna. “Obviously, China is producing world class research, but what the PLA and PLA-affiliated research institutions are doing specifically is much more difficult to discern.”

The July 1 death in a traffic accident in Beijing of a leading Chinese military AI expert provides a small window into the country’s ambitions.

At the time he died, Colonel Feng Yanghe, 38, was working on a “major task,” state-controlled China Daily reported, without going into detail. Feng had studied at the Department of Statistics at Harvard University, the report said.

In China, he headed a team that developed an AI system called “War Skull,” which China Daily said could “draft operation plans, conduct risk assessments and provide backup plans in advance based on incomplete tactical data.” The system had been used in exercises by the PLA, the report said.

The Biden Administration is so concerned about the tech race that it has moved to block China’s drive to conquer AI and other advanced technologies. Last month, Biden signed an executive order that will prohibit some new U.S. investment in China in sensitive technologies that could be used to bolster military capacity.

Anduril, the weaponry start-up created by VR-headset pioneer Palmer Luckey, has ambitions to be a major high-tech defense contractor. The Costa-Mesa, California-based company now employs more than 1,800 staff in the United States, the United Kingdom and Australia. Luckey’s biography on the company website says he formed Anduril to “radically transform the defense capabilities of the United States and its allies by fusing artificial intelligence with the latest hardware developments.”

Anduril said Luckey was unavailable to be interviewed for this story.

The core of Anduril’s business is its Lattice operating system, which combines technologies including sensor fusion, computer vision, edge computing and AI. The Lattice system drives the autonomous operation of hardware that the company supplies, including aerial drones, anti-drone systems and submarines such as Ghost Shark.

In its biggest commercial success so far, Anduril early last year won a contract worth almost $1 billion to supply U.S. Special Operations Command with a counter-drone system. The U.K. Ministry of Defense has also awarded the company a contract for a base defense system.

Arnott wouldn’t describe the capabilities of Ghost Shark. The vessels will be built at a secret plant on Sydney Harbour in close collaboration with the Australian Navy and defense scientists. “We absolutely can’t talk about any of the applications of this,” he said.

But a smaller, three-tonne autonomous submarine in Anduril’s product line-up, the Dive-LD, suggests what unmanned AI-powered subs can do. The Dive-LD can reach depths of 6,000 meters and operate autonomously for 10 days, according to the company website. The sub, which has a 3D-printed exterior, is capable of engaging in mine counter-warfare and anti-submarine warfare, the site says.

With no need for a pressure hull, Anduril's Dive-LD can descend far deeper than the manned submarines in military service. The maximum depths reachable by military subs is usually classified information, but naval analysts told Reuters it is somewhere between 300 and 900 meters. The ability to descend to much greater depths can make a sub tougher to detect and attack.

Veteran navy officers say dozens of autonomous submarines like Ghost Shark, armed with a mix of torpedoes, missiles and mines, could lurk off an enemy’s coast or lie in wait at a strategically important waterway or chokepoint. They could also be assigned to strike at targets their AI-powered operating systems have been taught to recognize.

Australia’s nuclear subsea fleet will be more formidable than the unmanned submarines of today. But, they will also take much longer to materialize.

In the first part of the project, the United States will supply up to five Virginia-class submarines to Canberra. The first of those subs will not enter service until early next decade. A further eight of a new class of subs will then be built starting from the 2040s, as part of the same AUD$368 billion project, under the AUKUS agreement, a defense-technology collaboration between Australia, Britain and the United States.

By the time this fleet is an effective force, big numbers of lethal robots operating in teams with human troops and traditional crewed weapons may have changed the nature of war, military strategists say.

“There is a lot of warfare that is dull, dirty and dangerous,” said Arnott. “It is a lot better to do that with a machine.”

Inspired by Ukraine war, Taiwan launches drone blitz to counter China https://www.reuters.com/investigates/special-report/us-china-tech-taiwan/

U.S. and China wage war beneath the waves – over internet cables https://www.reuters.com/investigates/special-report/us-china-tech-cables/

Inside the subsea cable firm secretly helping America take on China https://www.reuters.com/investigates/special-report/us-china-tech-subcom/

(Reporting by David Lague. Edited by Peter Hirschberg.)


The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.

More Related Articles

Info icon

This data feed is not available at this time.

Sign up for Smart Investing to get the latest news, strategies and tips to help you invest smarter.