Introduction of AI into the armed forces: experience of the USA, China and Israel

Experience of the USA, China and Israel in the use of artificial intelligence in the armed forces.

With the advent of AI systems, humanity has overcome another technological milestone, which has become a breakthrough, including in the military sphere. The potential capabilities provided by artificial intelligence (AI) are quite comparable to those of countries with nuclear weapons. However, at the moment, AI systems do not yet pose a threat similar to a nuclear apocalypse.

Unlike many technologies that have moved from the battlefield into everyday life, the path of AI has been rather the opposite. Although the current AI boom is largely driven by investment and research from the commercial sector, the military has been quick to recognize its potential. Its high adaptability and ability to quickly analyze large amounts of data in real time could not help but attract the attention of military structures.

Today, several key areas for integrating AI into military systems can be identified:

1. Autonomous and robotic combat systems. Creation and improvement of unmanned aerial vehicles, ground robots, sea drones and other platforms capable of performing tasks from reconnaissance to the direct use of force with a high degree of autonomy, minimizing direct human participation.

2. Intelligence, surveillance, reconnaissance and data analysis. AI is being adapted to process multimodal data (satellite imagery, intelligence information, media reports, etc.). Algorithms are capable of identifying hidden patterns, identifying targets and objects, and conducting terrain reconnaissance, providing command with an up-to-date operational picture.

3. Decision support and operations planning. Based on comprehensive data analysis, AI systems can predict the development of the situation on the battlefield, simulating the consequences of certain decisions.

4. Cyber warfare and information operations. AI is used to automate both defensive (searching for vulnerabilities and repelling cyber attacks) and offensive cyber operations. In addition, neural network deepfakes are used in information warfare to influence the enemy population.

This article presents a comparative analysis of the experience of introducing artificial intelligence technologies in the armed forces of Israel, the United States and China. The choice of countries is justified by the following factors: Israel, due to the ongoing military operations in the Middle East, has practical experience in the operational implementation and use of AI on the battlefield by its own armed forces; The United States maintains a dominant position in the IT sector, largely thanks to technology companies from Silicon Valley (as evidenced by the United States' first place in the Global AI Index, which evaluates countries based on key indicators such as the level of research and development, the state of infrastructure, the number of specialists, investments, government strategy and commercial adoption of technologies); China, which will invest between $84 billion and $98 billion in AI development in 2025, according to Bank of America, not only competes with the United States at a strategic level, but also represents an example of a successful state-led model in this area.

Israel: Operational Effectiveness in the Face of Asymmetric Threats

Israel is deliberately implementing a policy to integrate AI systems into its armed forces, seeking to compensate for the disproportion between the ever-growing arrays of intelligence data and the limited human resources for their operational analysis. From the earlier list of key areas for integrating AI into military systems, Israel has focused on developing AI systems for decision making and operations planning. The very logic of the development of Israeli military technologies based on AI suggests that Israel’s main goal in this direction is to qualitatively increase the efficiency of the command decision-making cycle. Analysis methods used before the “age of AI” can no longer cope with the deluge of data, which poses the risk of slower response to critical information. Thus, Israel is using AI to reduce the time between receiving data, identifying a target, and delivering a strike.

A key role in the development and implementation of AI systems in the Israeli armed forces is played by Unit 8200, created in 1952, part of the Military Intelligence Directorate of the Israel Defense Forces. It does not participate in combat operations on the ground, focusing primarily on cyber warfare.

Overall coordination of high-tech efforts (including AI) has historically been the responsibility of Israel's National Cybersecurity Authority. At the same time, developments are carried out in close cooperation with Israeli institutes (for example, the Technion, Tel Aviv University, the Weizmann Institute), commercial companies and, as it became known thanks to an investigation by the Israeli publications +972 Magazine and Local Call, Israeli engineers who were drafted into the army after the attack on October 7, who previously worked at Google and Microsoft.

According to data released by Israeli journalist Yuval Abraham, Israel routinely uses at least three AI-based systems: The Gospel/Habsora, Lavander and Where's Daddy? (Where is dad?). All of them were developed by the 8200 division.

"Gospel" automatically identifies and ranks targets for attack. These are mainly military facilities and infrastructure. Like all other systems, it was trained on decades of data collected by Israeli intelligence. “Gospel” is capable of quickly analyzing long-term arrays of raw data (satellite images, UAV data, communications interception, cyber intelligence data, thermal imaging images, etc.).

According to former IDF Chief of Staff Aviv Kochavi, “Gospel” allowed a radical increase in the number of targets: if before its implementation, the IDF identified about 50 targets per year, then after, during Operation “Wall Guard” (May 2021), “Gospel” generated up to 100 targets per day, using a unit of about 100 people.

"Lavender" became the next step in the evolution of "Gospel". It was created for the purpose of finding persons suspected of belonging to Hamas. The system, based on intercepted signals, including those obtained through Pegasus and similar software, and “indirect intelligence signs” (being in the same WhatsApp group with already known militants, frequent changes of place of residence or phone number, repetition of other patterns of behavior characteristic of Hamas members) assigns a rating from 1 to 100 to the suspect, reflecting the degree of probability of his belonging to Hamas, and also marks his location. According to an investigation by +972 Magazine and Local Call, in the first weeks after the attack on October 7, 2023, the system included up to 37 thousand Palestinians living in Gaza as potential targets.

The use of Lavender by the Israeli army is accompanied by accusations of war crimes. The expert community indicates a high risk of error when choosing a target for elimination in a constantly changing situation on the battlefield and the need to quickly make decisions. In particular, it is alleged that at the beginning of the war, the target for destruction was approved after a superficial check by the operators. About 20 seconds were allotted for selecting and assessing each target, this was only enough to make sure that the found target was a man, since there are no women among the Hamas military wing. This approach, which allows for a margin of error of more than 10%, has dramatically increased the number of indiscriminate strikes, with collateral casualties among women and children.

System "Where is daddy?" closely integrated with Lavender. It is designed to track in real time the geolocation of a target previously identified by Lavender. “Where is dad?” calculates when the target returns home and determines the time window for striking. Journalists confirm frequent cases when attacks were carried out on residential buildings, which resulted in the death of not only militants, but also all members of their families.

Drastic changes in the use of AI systems in combat occurred during Operation Iron Swords. The political decision was to recognize the target for the liquidation of any member of the military wing of Hamas, regardless of his position in the hierarchy. The number of targets increased multiple times, which made it impossible for them to be rechecked qualitatively by a human operator.

Unit 8200's developments represent some of the most advanced, but also the most legally and ethically controversial military programs in the world. International humanitarian law is based on the principles of distinction, proportionality and precaution in the conduct of hostilities. Systems like "Lavender" and "Where's Dad?", which identify combatants through algorithms with a greater than 10% error rate by marking their location in residential buildings, clearly violate all of these principles, adding to the arguments of human rights activists who condemn Israel's methods of warfare. Israel's refusal to sign the 2023 Hague REAIM Declaration on the responsible use of AI in the military sphere was a clear signal that Israel will not accept restrictions in this area, although this goes against the opinion of the international community and moral and ethical standards.

At the same time, the use of AI systems in high-intensity operations, such as in Gaza, is fully consistent with Israel’s own military strategy - the Dahiya Doctrine, which involves the use of massive, disproportionate force, the destruction of civilian infrastructure in order to demoralize the enemy. In line with the Dahiya Doctrine, the IDF plans to integrate AI into half of its military systems by 2028, including further automation of reconnaissance and fire control.

USA: public-private symbiosis and global leadership

The US Department of Defense first published a strategy governing the use of AI for military purposes in 2018. The document stated that AI technology is key to “fighting and winning future wars.” Currently, the document defining the US national strategy in the field of AI development is US Presidential Executive Order No. 14101, “Removing Obstacles to American Leadership in Artificial Intelligence,” dated January 23, 2025. Subsequently, its implementation was specified in the action plan “Winning the AI Race: The US AI Action Plan,” published by the White House on July 23, 2025. It emphasizes the need overcoming bureaucracy, increasing investment and strengthening cooperation between the public and private sectors. These documents provided government support for rapidly developing AI technologies in the United States, including those commercial developments that could be used on the battlefield.

The most ambitious public-private initiative was the Stargate project, announced in January 2025 by US President Donald Trump. The project brings together OpenAI, SoftBank, Oracle and investment firm MGX to create a network of advanced data centers for data processing and AI training in the United States by 2029.

The main beneficiary of government investments in AI remains the military department. Thus, about 75% of US government spending on AI-related projects from 2013 to 2023. fell on the Pentagon. Coordination of US efforts in the field of military AI falls on several key structures:

1. Defense Advanced Research Projects Agency (DARPA). Created in 1958 in response to the USSR's launch of the first artificial Earth satellite, it has now become a key Pentagon department responsible for the development and introduction of new technologies into the US military.

2. Defense Innovation Unit (DIU). Based in Silicon Valley, he is responsible for the Pentagon’s contacts with IT startups. DIU engages the private sector in defense innovation projects, and essentially serves as a “bridge” between the Pentagon and the dynamic AI sector.

3. National laboratories. Research centers such as the Lawrence Livermore National Laboratory. E. Lawrence (Lawrence Livermore National Laboratory, LLNL) or Los Alamos National Laboratory (LANL), known for developing nuclear weapons in the middle of the last century, and today they are also actively involved in the development of AI systems, including dual-use ones. Some of these developments are carried out in close contact with IT startups.

Many of America's advanced military AI systems have civilian roots. It has become an established practice to develop a civilian prototype in the first stage, and then adapt it to military needs in the second. For example, the Microsoft HoloLens headset (including HoloLens 2 and accompanying Azure cloud services) was initially developed as a commercial AR product for training, remote assistance and working with 3D models, and then technical developments and cloud infrastructure became the basis for the development of the Integrated Visual Augmentation System program, an augmented reality system for the US Army.

In a sign of strengthening private sector cooperation with the Pentagon, four executives from major technology companies have joined the U.S. Army Reserve. In June 2025, the following took the oath and became lieutenant colonels: Shyam Sankar (Palantir), Andrew Bosworth (Meta, recognized as extremist and banned in Russia), Kevin Weil (OpenAI), Bob McGrew (Thinking Machines Lab and OpenAI). Top managers headed the new military unit Detachment 201 (“Detachment 201”).

It is expected that leaders of leading companies in the IT sector will engage in attracting people with technological experience to work for the benefit of US national security. IT reservists will serve 120 hours per year, have flexible schedules and are exempt from military training.

All these events signaled a paradigm shift in the relationship between the government and the technology sector. Just ten years ago, IT companies avoided working on technologies that could be used for military purposes. Now, for example, Google and Open AI are lifting internal protocols banning the use of their AI developments in weapons production, paving the way for closer cooperation with the defense industrial sector.

This strategic turn found its practical application in hot spots. The Ukrainian conflict has become a testing ground for American military AI systems. From the very beginning of the SVO, heads of IT startups from Silicon Valley began to come to Ukraine. One of the first was the CEO of Palantir Technologies Alex Karp. This company has created the MetaConstellation platform, which models a comprehensive digital battlefield. It integrates and analyzes data from satellites including Starlink, Maxar, Airbus, ICEYE, Capella and NOAA, etc., reducing the detection of concentrations of troops, equipment or artillery positions to 2-3 minutes.

Other American startups that have proven themselves in Ukraine include Primer AI and Clearview AI. Primer AI, which specializes in natural language processing and machine translation. Primer AI's developments make it possible to overcome the language barrier, decipher, denoise and translate intercepted tactical radio communications. Clearview AI has created a biometric image identification system with a claimed recognition accuracy of 99%. The Clearview AI database contains more than 40 billion facial images (during the SVO it was increased by 400%), collected from open sources - profiles on the social networks Facebook, Instagram (owned by Meta, which is recognized as extremist and banned in Russia), VKontakte. The group Privacy International states that the Ukrainian Armed Forces are using this system to identify dead and captured Russian military personnel.

General guidance on working with American AI startups that are ready to provide their systems for the Armed Forces of Ukraine is provided by the Minister of Digital Transformation of Ukraine, Mikhail Fedorov. He openly talks about testing foreign technologies in Ukraine, with an emphasis on testing innovations on the battlefield in real time. “We are ready to help companies from partner countries develop, test and refine technologies that actually work on the battlefield. This is a chance to gain experience that simply cannot be simulated in the laboratory,” the Ukrainian minister emphasized.

The beginning of the SVO coincided with the global rise of the AI industry, and many American AI startups are offering their developments to the Ukrainian army on a pro bono basis, not only for practical testing, but also, probably, to create a reputation as a company whose developments were used in military operations.

The Pentagon is also investing heavily in autonomous systems and agent-based AI processes, focusing on systems that can perform tasks with minimal human intervention. The following areas attract particular attention:

·         Autonomous drones. The Replicator program has been underway since 2023, with the goal of deploying thousands of autonomous drones by 2026. To implement this program, the Pentagon entered into contracts with Shield AI ($167 million) and Anduril to improve swarm control algorithms.

·         Collaborative combat aircraft. The US Air Force is systematically introducing combat wingman drones operating in conjunction with manned F-35 fighters.

The United States' efforts to introduce AI into its military are driven by the need to get ahead of China, which is seen by the United States as a major competitor at all levels. The US National Security Commission on AI, founded in 2021, predicts that China could become a leader in AI within ten years.

These concerns are based on the fact that China is investing more in projects related to military AI than the United States. The exact figures for China's military spending on AI are classified, but a study by the US Center for Security and Emerging Technologies (CSET), analyzing open PLA contracts, estimated the minimum level of Chinese purchases of systems with AI elements at $1.6 billion annually. The actual investment may be significantly higher.

China: state-managed integration and total intellectualization

China, recognizing the high transformational potential of AI, is betting on creating an intellectualized military force. His approach is also based on the policy of military-civil integration, but government control over the development of military AI systems is incomparably stricter than in the United States. In China, the development of military AI is directly controlled by the state through the civil-military integration policy, which obliges private companies to cooperate with the army, and centralized five-year planning. Thus, the 14th Five-Year Plan, implemented from 2021 to 2025, directly instructs the PLA to “accelerate the comprehensive development of mechanization, information and intellectualization” of AI systems. While in the US private companies like Google retain significant autonomy and can challenge military contracts under public pressure, Chinese companies are legally subject to party directives. In addition, China's Data Security Law gives the government direct access to company data, whereas in the United States such practices are limited by judicial and public oversight.

Development of dual-use AI systems began around the 1980s with the launch of Plan 863, which included a wide range of intelligent robotics projects. Today this is done by the National Key Research and Development Programme, which funds machine learning projects. The implementation of China's military AI strategy is ensured through close coordination of military and civilian structures, including:

1. PLA Academy of Military Sciences.

2. State defense corporations. For example, Nornico (Northern Industry Corporation of China) designs autonomous systems, in particular, the Intelligent Precision Strike System (a complex for regulating a swarm of drones and simulating the battlefield), and China Electronics Technology Group Corporation is involved in the development of AI systems for electronic warfare, communications, interception and signal processing.

3. Private technology companies attracted as part of the vector of military-civil integration. Baidu produces autonomous driving and cloud computing technologies. iFLYTEK specializes in speech recognition and natural language processing. The products are used by the PLA to automate eavesdropping and interception of communications. PIESAT sells geospatial data derived from AI-powered feature recognition, which is used for real-time location mapping, and also provides satellite monitoring and data analysis services for surveillance and reconnaissance.

The PLA is deploying AI in a wide range of military applications:

1. Autonomous and unmanned systems.

China is developing highly autonomous attack UAVs ("Dark Sword", "Star Shadow", "Sharp Sword") with the ability to fly autonomously, identify targets and use weapons. Work is underway on UAV swarm technology.

In addition to flying autonomous systems, China is working on land and sea drones. One of the first functioning ground robots was shown in exercises in Cambodia back in 2021. “Robot Dogs” demonstrated the ability to carry and use autonomous small arms. Among the maritime systems, unmanned boats such as Jinghai stand out for autonomous patrol of the waters and unmanned corvettes D3000 with surface combat functions.

2. Command, communications, intelligence and cyber operations.

The PLA Strategic Support Force uses AI for cyber defense, including pattern recognition to defend against attacks and detect network intrusions. In addition, active work is underway to create AI systems to collect data from various sources and speed up decision-making. The main goal is to create an “all-weather multidimensional situational awareness network.”

It is worth noting that in parallel with China, the United States is designing a similar network that connects various platforms and domains under the concept of “Joint Agency Command and Control” (JADC2).

3. Simulations and exercises.

China is developing digital twins and intelligent simulators for education and training. Thus, in May 2024, the journal Common Control & Simulation presented the “Virtual Commander” AI system, which simulates the decision-making style of PLA commanders in computer battle simulations. In the process, she demonstrates the ability not only to quickly analyze the situation, but also to generate options for action, assessing their effectiveness.

With its active development of AI, China is advocating for the responsible and regulated use of military AI. In 2021, China presented a document at the UN on “Regulating the Military Use of Artificial Intelligence,” and in 2023 it put forward a global Artificial Intelligence Governance Initiative, which was not publicly supported by the United States and Israel, but was approved by many countries in the global South. In these documents, China proposes the following provisions that combine technological development and ethical standards: maintaining control over humans, observing international humanitarian law in the development and use of AI in the military sphere, and preventing an arms race in the field of AI.

Comparative analysis

The analysis allows us to identify three directions of militarization of AI. Israel has chosen a narrowly focused model, focused on solving specific problems at the cost of neglecting the norms of international law. The United States is pursuing an open innovation model, backed by the power of the private sector. China remains committed to a socialist market economy, where the state directs the efforts of all sectors to introduce high technology into the armed forces, while developing national standards.

The experiences of Israel, the United States and China demonstrate that the introduction of AI systems into the armed forces is fraught with significant challenges. The comparative analysis identified the following problems that AI developers face everywhere:

1. Lack and/or unavailability of relevant data. The main difficulty is the lack of sufficient data directly related to military operations. Much of the data critical to training AI systems is either not collected digitally (e.g., recorded on paper) or is classified.

In addition, the issue of data fragmentation is acute, when the exchange of information between different branches of the military is extremely difficult. This creates “information islands”, preventing the creation of a unified training base.

2. Vulnerability of young AI systems. Many experts recognize current AI systems as highly vulnerable to targeted cyber attacks and sabotage. Attackers can manipulate data and algorithms during the training phase of the system. In particular, “poison” the training data by introducing hidden distortions, which leads to systematic errors in the operation of the system, the root of which will lie in the algorithm itself. Additional complexity is created by the large volume of data, which is difficult to protect. Experts note the problem of the inability to reliably detect the fact of an enemy’s penetration into one’s own networks.

3. Network restrictions. The modern battlefield generates a colossal amount of information: images, video from drones, etc. Existing communication channels are not always capable of providing real-time transmission. Even with sufficient bandwidth, delays in data transmission reduce responsiveness.

Particular difficulties are created by the specifics of the media: air and water. Thus, creating a reliable network for a swarm of UAVs that requires constant exchange of information between many “nodes” is more difficult than controlling single drones over a point-to-point channel. Communication underwater is even more limited due to signal absorption. In addition, both UAVs and underwater autonomous vehicles often rely on ground or airborne relays, which are vulnerable to missile attacks.

4. Difficulties in testing and evaluation. To establish trust in AI systems, especially for military purposes, comprehensive testing is necessary to confirm their combat effectiveness. At the same time, it is obvious that it is fundamentally impossible to test autonomous weapons systems in all possible scenarios of a real war. Such modeling requires time and financial resources, and still cannot be exhaustive. The very process of testing autonomous systems, especially in conditions close to combat, may pose safety risks.

In general, despite significant differences in strategic approaches, the experience of Israel, the United States and China in introducing AI into the armed forces reveals a number of universal technological and operational trends that are global in nature and determine the general trends in the development of the military industry.

Firstly, a focus on creating integrated decision support systems. Israeli developments “Gospel” and “Lavender”, American programs like Palantir’s MetaConstellation and Chinese projects to create an “all-weather multi-dimensional situational awareness network” demonstrate a general paradigm of the transition from a human analyst to a human operator monitoring algorithms.

Secondly, the development of autonomous and swarm combat systems. This trend reflects the global search for demographic constraints through the delegation of tactical functions to autonomous systems.

Thirdly, the emergence of developed forms of military-civil integration. Thus, the United States has created institutional “bridges” (DIU, “Unit 201”) that attract civilian startups, China is directively integrating private companies into defense programs, and Israel is using a pool of developers from tech giants. The methods are different, but the essence remains the same: no country can develop military AI in isolation from the civilian technical sector, which creates new partnerships between private IT companies and the army.

All three directions set the following vectors for the development of modern military-technical policy: total digitalization of intelligence, delegation of functions to AI systems while maintaining control over humans, and institutional coupling of military departments with the IT sector. The experience of Israel, the United States and China shows that the future of military superiority is determined by the ability to quickly implement this triad of interconnected changes, rather than simply focusing on individual breakthrough technologies.