AI: Seek and Destroy

The Autonomous Weapons Industry

By Laura Carrer, Davide Del Monte, Andrea Daniele Signorelli
Edited by Lorenzo Bagnoli
October 2026

Dedicated to Carola Frediani

Mapping the Infrastructure of Algorithmic War

This research addresses a fundamental shift in modern warfare: artificial intelligence is no longer futuristic speculation, but an active, widespread reality on the battlefield.

We have moved well beyond the neat confines of Asimov’s First Law—which states that “a robot may not injure a human being or, through inaction, allow a human being to come to harm”. Instead, we are entering a second “Oppenheimer Era.” In this new global arms race, the breakthrough lies not just in the sheer firepower available to armed forces, but in a machine’s ability to analyze context and make probabilistic decisions about whom, where, and when to strike.

This book aims to provide a comprehensive map of the complex network underpinning autonomous weapons—from research and development to manufacturing and sales. It provides journalists and researchers with the analytical framework needed to track the sector. This emerging market is dominated by opaque actors.

The five chapters examine the rise of autonomous weapons systems and the resulting transformation of modern warfare through multiple lenses.

Chapter One traces the origins and definitions of Lethal Autonomous Weapon Systems (LAWS), tracking their evolution from early rudimentary torpedoes to fully AI-guided platforms. In recent years, these systems have progressed from basic target identification to autonomous decision-making. AI has effectively climbed the military command chain—promising to lift the “fog of war” while introducing unprecedented ethical risks.

Chapter Two turns to the battlefields of Ukraine, Gaza, Iran, and Lebanon, which serve as testing grounds for integrated surveillance, cloud infrastructure, and targeting algorithms. We examine controversial intelligence platforms like Lavender and Gospel—described by critics as automated assassination pipelines—where human oversight is often reduced to a multi-second rubber stamp.

Chapter Three examines Europe’s emerging “defense tech” ecosystem, where private capital and NATO funds are flooding the sector as legacy investment taboos are discarded. We look at new defense “unicorns” like Helsing and ICEYE, which offer high-performance technologies aimed at bolstering continental autonomy.

Chapter Four explores the concept of a “new Leviathan”—the growing convergence of resource extraction, big tech, and the military apparatus. We trace the supply chain from raw material extraction to the AI systems that detect, track, and strike targets. This chapter contrasts traditional institutional funds—which act as the backbone of the sector—with a new wave of fast-moving startups backed by risk-tolerant venture capital.

Chapter Five analyzes European Union support for autonomous systems through the European Defence Fund (EDF), alongside the central and controversial role played by public universities. As the line between civilian and military tech blurs, the militarization of academic research is sparking intense ethical debate on campuses and testing institutional independence.

Ultimately, beyond documenting new technology and its obvious ethical dilemmas, this book uncovers the underlying structural forces reshaping the global balance of power—and the immense risks they pose to us all.

The Specter of Autonomous Weapons

by Andrea Daniele Signorelli

Why the growing urgency surrounding autonomous weapons? Some analysts¹ and military experts argue that AI-powered lethal autonomous weapon systems (LAWS) are merely the logical, "smart" evolution of arms that have been on the battlefield for decades. This argument is designed to manage public concern by framing these systems not as a novel threat, but as a major leap forward—machines capable of distinguishing civilians from combatants with far greater precision than traditional anti-personnel mines, thereby minimizing collateral damage.

Is that really true? In one narrow sense, yes: autonomous weapons are not fundamentally new. As early as 1979, the U.S. Navy unveiled the Mark 60 CAPTOR, a seabed-anchored anti-submarine mine fitted with a torpedo that fired automatically when onboard sensors confirmed a designated target within range. Nor was that the earliest instance of partial autonomy. In 1943, Nazi Germany deployed the G7es T5 Zaunkönig acoustic torpedo, which tracked the sound of ship propellers after launch. The 1950s brought the U.S. AIM-9 Sidewinder (1956), an infrared-guided air-to-air missile that locked onto heat signatures independently once airborne. By 1978, the U.S. Navy had fielded the Phalanx CIWS, designed to automatically detect, track, and destroy incoming anti-ship missiles.

Then there are traditional anti-personnel landmines—unmanned weapons in use for over a century that detonate upon contact, requiring no human operator to execute the strike. These examples show that rudimentary forms of autonomous or semi-autonomous hardware have been around for a long time.

Yet focusing exclusively on autonomy at the moment of detonation misses the central point. Modern AI-driven technology—such as computer vision models on drones that identify vehicles, infrastructure, or personnel, or algorithms processing satellite intelligence to flag targets—does not simply trigger a fixed action in response to a hardcoded physical stimulus like pressure or heat. Instead, these systems evaluate complex data to make contextual choices. Autonomy has shifted from how a weapon is triggered to who, where, and when to strike. This fundamental shift—and its far-reaching implications—gives rise to today’s ethical, operational, and strategic challenges.

What, then, defines an autonomous weapon? Current official definitions offer little clarity. Even within the UN Convention on Certain Conventional Weapons (CCW)², an autonomous weapon is defined broadly as a system that, once activated by a human, can select and engage targets without further human intervention. Under this framework, a 1970s Mark 60 CAPTOR and a modern AI target-selection system fall into the exact same category. The definition fails to capture the qualitative leap brought by artificial intelligence, which moves beyond executing preset mechanical rules to classifying, evaluating, interpreting, and making probabilistic target assessments in fluid war zones.

The crucial distinction is between weapons that activate autonomously and weapons that autonomously determine targets and engagement conditions. A 2023 U.S. Department of Defense policy document outlines three distinct operational levels³: an autonomous weapon system can select and engage targets without operator intervention once activated; an operator-supervised autonomous weapon system gives a human the ability to monitor operations and intervene or abort in the event of a malfunction; and a semi-autonomous weapon system engages only those targets explicitly chosen by a human operator, even if the system autonomously navigates the area, acquires the target, and calculates fire parameters.

In practical human terms, this three-tier model translates to human-out-of-the-loop (no monitoring or intervention possible), human-on-the-loop (automated execution with human oversight and abort capability), and human-in-the-loop (human retains final authority over target selection)⁴. Autonomy in AI-driven warfare is therefore best understood not as an on-off switch, but as a spectrum reflecting varying degrees of delegated decision-making authority. As The Guardian noted, experts and researchers broadly agree that machines are now more capable of making independent battlefield decisions than at any point in history⁵.

Examples of AI-Powered Autonomous Systems

In popular culture, the sci-fi archetype of the humanoid "Terminator" robot has largely given way to a more realistic threat: autonomous drones that use AI to navigate terrain, identify targets, and execute strikes. The most prominent example is the Kargu, developed by Turkish defense firm STM⁶—a loitering munition built for reconnaissance, surveillance, and precision strikes. According to STM, while the Kargu can be piloted manually, it can also operate fully autonomously using onboard computer vision, target recognition algorithms, and real-time machine learning models to strike fixed or moving targets.

However, the most technologically ambitious projects are emerging from the U.S. defense startup Anduril. Founded in 2017 by Palmer Luckey (creator of Oculus) and backed by major venture capital firms including Founders Fund, Andreessen Horowitz, General Catalyst, Lux Capital, and Altimeter, Anduril has seen explosive growth. A 2026 funding round led by Thrive Capital and Andreessen Horowitz pushed total capital secured to $10 billion, valuing the company at roughly $60 billion. Government spending has surged alongside private investment: after securing $1.5 billion in defense contracts in 2024, the U.S. Army awarded Anduril a 10-year contract worth up to $20 billion in March 2026⁷.

Despite these massive figures, many of Anduril's marquee platforms remain under development or in prototype testing. Notable deployed systems include Sentry Towers—autonomous surveillance masts equipped with radar, optical sensors, and AI software to track and classify movements along borders; the Dive-LD, an autonomous long-range underwater vessel designed for seabed reconnaissance and intelligence gathering; the Anvil and Roadrunner anti-drone interceptors, which destroy hostile aircraft via kinetic impact; and Altius loitering munitions—drones that loiter over a designated zone before diving directly into a target.

Anduril is far from alone in commercializing AI for the battlefield. The Pentagon requested over $13 billion from Congress for autonomous systems development in its latest budget⁸. A substantial portion of those funds is earmarked for contracts with specialized tech startups—such as Shield AI, Saronic, and Epirus—as well as legacy defense primes like Lockheed Martin, L3Harris, and Northrop Grumman.

The world's other tech superpower, China, is matching this push. At the 2024 Zhuhai Airshow, state defense conglomerate Norinco displayed networked ground armor and drone swarms controlled and coordinated via central AI networks. At the same defense expo, the Aviation Industry Corporation of China (AVIC) unveiled a 16-metric-ton platform designed as an airborne mothership capable of launching dozens of smaller swarming drones mid-flight. More recently, Norinco introduced the P60, an autonomous unmanned ground combat vehicle built for reconnaissance and fire support that can integrate with remote weapon stations.

While public details on Chinese programs remain limited, defense analysts—including Emelia Probasco of the Center for Security and Emerging Technologies—note that the People's Liberation Army is not only matching U.S. technical capabilities but may hold a distinct edge in operational scale due to its superior manufacturing capacity⁹.

Elsewhere, Israeli firm Elbit Systems produces the Assault Rifle Combat Application System—an AI-enabled rifle sight designed to detect and identify human targets at ranges exceeding 500 meters, embedding algorithmic targeting directly into infantry small arms. In South Korea, Samsung developed the automated SGR-A1 sentry gun in the 2000s for the Demilitarized Zone; equipped with thermal sensors and a heavy machine gun, the fixed turret can track and potentially engage human targets (though its exact degree of lethal autonomy remains classified). South Korean defense firm DoDAAM manufactures the Super aEgis II, a similar automated turret capable of identifying human targets several kilometers away.

This global arms race extends to Russia's ZALA/Kalashnikov Lancet loitering munitions, first-person view (FPV) strike drones built by Ukraine’s TAF Industries, and a rapidly expanding ecosystem of European defense tech startups. As The Guardian observed in a direct comparison to the Manhattan Project, the race for autonomous weapons "has reached its Oppenheimer moment."¹⁰

As Eric Schmidt—former Google CEO and former chair of the U.S. National Security Commission on Artificial Intelligence—observed: "Every so often, a new technology or a new weapon emerges that changes the game. In the 1930s, Einstein wrote a letter to Roosevelt¹¹ explaining that there was a new technology—nuclear weapons—that would change the nature of war, as indeed it did. I believe that AI-powered warfare systems are just as powerful."¹²

Yet while the strategic implications of nuclear weapons—their absolute destructive force and geopolitical deterrent value—are well understood, the capabilities, failure modes, and long-term consequences of AI-driven warfare remain largely uncharted. Before examining those risks, however, we must look at how artificial intelligence has moved up the military chain of command—shifting from tactical kinetic weapons to high-level Decision Support Systems (DSS), amplifying concerns over algorithmic error, civilian risk, and moral accountability.

BOX: TAXONOMY OF AUTONOMOUS WEAPONS

From Kinetic Weapons to Strategic Decision Tools

As artificial intelligence matures, its primary center of gravity in modern warfare is shifting from autonomous physical hardware to algorithmic Decision Support Systems (DSS). This evolution underscores the fundamental shift noted earlier: AI autonomy is no longer confined to triggering a physical weapon, but now governs the high-level decision of whom, where, and when to target. The task of patrolling territory and identifying threats is increasingly handled not by drones, ground vehicles, or naval craft, but by software platforms engineered to ingest live video feeds, process high-resolution imagery, perform predictive threat analyses, model potential collateral damage, and flag behavioral anomalies in real time.

Inside a modern AI-enabled command center, operators do not look at raw video feeds; they monitor dynamic digital maps populated by hundreds of moving icons: friendly troop formations, suspected enemy positions, loitering strike drones, supply convoys, and intercepted radar tracks. This operational picture is compiled by aggregating vast, disparate data streams: military and commercial satellite imagery, live drone feeds, ground-based radar and thermal telemetry, intercepted enemy communications, human intelligence (HUMINT) reports, mobile phone geolocation data, and open-source intelligence (OSINT) scraped from social media.

Without automated decision support, this torrent of raw intelligence would overwhelm analysts, creating a severe operational bottleneck. A DSS filters, correlates, and synthesizes these complex inputs into actionable military options. Consider a typical operational scenario: the software flags an enemy convoy along a specific transit route, calculates an 87% probability that the vehicles are heavy armor, and instantly generates three recommended Courses of Action (COA)—an immediate airstrike, extended aerial surveillance, or a request for manual intelligence verification. Operators do not compile or cross-reference this data themselves; they receive a pre-structured analysis, allowing them to authorize strikes in seconds.

In essence, these DSS platforms perform two critical functions: dispelling the classic "fog of war"—the operational uncertainty famously conceptualized by Prussian military theorist Carl von Clausewitz—and dramatically accelerating the command decision cycle. By synthesizing satellite, sensor, and intelligence data at machine speed, a DSS evaluates operational choices, weighs collateral risks, coordinates tactical units, and adapts to battlefield developments in near-real time.

The flagship example of an enterprise-grade Decision Support System is the Maven Smart System. Originally developed by Palantir beginning in 2017 under the Pentagon's Project Maven, it functions as a comprehensive command-and-control (C2) platform. Maven delivers real-time situational awareness—a live digital display of friendly and hostile forces, available combat assets, and active threats—while actively assisting operational planners by generating action plans and drafting military orders. Following NATO’s 2025 contract with Palantir to deploy the platform, alliance officials noted that Maven provides commanders with "video-game-style oversight" of active war zones.

In recent years, military decision support platforms like Maven—as well as specialized target-generation software like Israel's "The Gospel," which processes mass intelligence data to auto-generate lists of target facilities, such as buildings, depots, and tunnels—have begun integrating Large Language Models (LLMs) such as OpenAI’s GPT and Anthropic’s Claude¹³.

This fusion pairs predictive AI (built to analyze massive datasets, extract structural patterns, classify entities, and calculate probabilities) with generative AI (capable of parsing natural language queries and producing executive summaries, briefing notes, and operational recommendations). By embedding LLMs into predictive military platforms, defense developers aim to make complex data synthesis intuitive for human commanders. As an analytical paper published by NATO’s Joint Air Power Competence Centre noted:

"LLMs can assist military commanders in making rapid, high-stakes decisions by helping staff draft, evaluate, and prioritize available Courses of Action (COAs). By supporting human operators in processing real-time intelligence, LLMs compress the operational decision cycle, providing a decisive advantage on the battlefield."¹⁴

In short, platforms like Maven aggregate satellite imagery, drone telemetry, intelligence logs, and radar tracks. Integrated LLMs like Claude or ChatGPT then translate that complex data mosaic into plain language, recommending specific strikes or force deployments. Rather than replacing predictive models, generative AI complements them within the same digital ecosystem—radically accelerating how military command structures analyze, digest, and act upon battlefield intelligence.

Ethical and Security Issues

Autonomous weapons and decision-support systems share many of the same core vulnerabilities: algorithmic opacity, ambiguous chains of liability, human overreliance on automated judgment, the progressive erosion of human decision-making skills, and the lowered threshold for kinetic escalation. Yet significant operational distinctions remain¹⁵.

The hazards of autonomous weapons manifest primarily at the point of lethal execution. Because the physical machine carries out the strike, the key question is whether it can comply with International Humanitarian Law—accurately distinguishing civilians from combatants, identifying subtle signs of surrender, reading contextual cues, and aborting strikes when collateral risk to civilians becomes excessive.

Decision Support Systems (DSS), by contrast, do not fire kinetic rounds; they generate intelligence summaries, operational recommendations, and target lists. While the final decision formally rests with a human commander, it is made under conditions orchestrated by the software. With a DSS, the central problem is not whether a machine kills in place of a human, but whether the human operator can meaningfully exercise choice when forced to digest automated outputs generated at a volume and velocity that preclude informed decision-making.

In light of AI's propensity for error, human oversight remains vital. Yet when operational speed is prioritized, "meaningful human oversight"—a term now central to international arms control negotiations—risks becoming an empty formality. Human cognitive latency acts as a bottleneck to software speed, threatening to transform DSS from decision-support tools into de facto decision-making systems. Furthermore, human capacity to supervise, correct, or override algorithmic recommendations is routinely overstated.

This human failure mode is well-documented. A study from the Wharton School identifies "cognitive surrender"¹⁶ as the phenomenon where users of large language models spend progressively less effort verifying output quality. This represents an insidious variant of classic "automation bias"—the tendency to trust automated systems due to an exaggerated perception of algorithmic objectivity and uncritical assumptions about machine capability.

An additional emerging hazard is military de-skilling driven by accelerated operational tempos, which forces command structures to delegate increasing authority to software. "We are diminishing our own skills," notes Elke Schwarz, a professor in the Department of War Studies at the University of London. "Commanders are becoming less and less capable of identifying what they are responsible for on the battlefield."¹⁷

This dynamic is further compounded by "AI sycophancy"—the documented tendency of models to flatter user preferences. Inverting automation bias, where humans trust the machine too implicitly, sycophancy occurs when the machine tells the human what they want to hear. On the battlefield, operators risk leveraging this dynamic—consciously or unthinkingly—to obtain algorithmic validation for flawed, pre-existing decisions.

Then there is the persistent hazard of AI hallucinations—where models present fabricated data as objective fact within intelligence summaries, briefs, or translations. This risk is heightened by recent studies showing generative models consistently favoring aggressive military escalation over diplomatic or defensive choices in combat simulations¹⁸.

These failure modes directly impact questions of ethics, accountability, and transparency. How can liability be established along a complex chain of command when defense establishments, such as the U.S. military¹⁹, frequently do not disclose whether AI tools were deployed in a specific strike? And where does legal liability ultimately lie: with the soldier who executed the strike, the officer who validated the algorithmic target, the commander who deployed the software, or the software vendor itself?

In current practice, commercial vendors remain largely insulated from liability. Defense contractors typically hold liability protections when proprietary platforms fail in the field, even with lethal consequences. Combined with government security classification, neither vendors nor militaries are required to disclose underlying software mechanics²⁰. As University of Virginia legal scholar Ashley Deeks observes, blending proprietary commercial software with classified national security architecture creates a "double black box"²¹ that shields algorithmic operations from public and legal oversight.

Conversely, Carnegie Mellon researcher Carol J. Smith²² argues that accountability should not fall solely on the lowest-ranking operator or the top commander, but should be distributed across all stakeholders involved in the system's development and deployment. From this perspective, maintaining a human-in-the-loop and preserving transparent decision logs ought to resolve liability disputes.

Yet this framework may be overly simplistic. As Paul Scharre, Executive Vice President of the Center for a New American Security (CNAS), observes, defense establishments often retain a "human-in-the-loop" primarily to ensure someone remains legally blameworthy when things go wrong.

Ultimately, the core issue remains unchanged. Autonomous weapons have existed for decades; focusing exclusively on sci-fi depictions of "killer robots" obscures the true qualitative leap introduced by artificial intelligence: the capacity to interpret context, classify targets, evaluate options, and make probabilistic decisions. This capability powers both autonomous strike drones and the command software now advising officers on whom, where, and when to strike. It is within this gray area—where humans retain formal command authority while functioning primarily as rubber stamps for machine choices—that the most urgent ethical, political, and legal battlegrounds are taking shape.

Autonomous Weapon Systems in Recent Conflicts

by Laura Carrer

In recent years, autonomous weapon systems have evolved from a theoretical concept into a defining feature of modern warfare. Conflicts across Ukraine, Gaza²³, and, to a lesser extent, Iran and Lebanon show an increasing convergence of sensor arrays, surveillance infrastructure, decision support software, cloud networks, and automated data analytics, representing a broader shift toward remote, algorithmic, and data-driven armed violence²⁴.

In this landscape, the core concern is not merely whether fully autonomous weapons are currently deployed on a massive scale, but the degree to which incremental automation is systematically shrinking the scope of meaningful human control²⁵ over target selection and engagement.

The primary conceptual hurdle is one of definition. As Human Rights Watch highlights in its 2025 report on autonomous weapons systems and digital decision-making, the legal and political debate focuses on systems that, once activated, select and deliver kinetic force against targets based on real-time sensor processing. This mechanism fundamentally distinguishes autonomous weapons from both remotely piloted drones and software platforms that merely assist human decision-makers without directly employing kinetic force.

This distinction is essential: much of what is currently observed on the battlefield occupies a murky gray zone between advanced automation, semi-autonomy, and the systemic handoff of cognitive and operational functions to algorithmic architectures.

Analyzing these dynamics requires a critical framework like the one developed by Dr. Lauren Gould at Utrecht University through the Intimacies of Remote Warfare and Realities of Algorithmic Warfare research initiatives. These programs examine the structural impacts of remote warfare—the contemporary "Western way of war"—focusing on civilian harm, democratic transparency, and mechanisms of accountability.

Autonomous weapons are not an isolated innovation; they represent the latest iteration of a long-standing trajectory of remote warfare. In this evolution, geographical distance, technological superiority, and data mediation continuously redefine the relationship between violence, legal responsibility, and public visibility²⁶.

As Dr. Gould noted in an interview with Follow the Money, contemporary warfare increasingly relies on satellites, uncrewed aerial vehicles (UAVs), surveillance grids, and AI platforms capable of tracking and striking targets with unprecedented speed and range. Yet the primary issue extends beyond the physical distance between the operator and the target area. Rather, this distance degrades contextual human intelligence, forces an over-reliance on remote sensor feeds, and obscures the decision-making processes governing the use of force.

The result is a mode of warfare that conveys an impression of surgical precision and rationality, yet frequently inflicts catastrophic harm—especially when underlying data is flawed, decontextualized, or corrupted.

From Kyiv and Gaza to Tehran

Applying this analytical framework across active operational theaters—spanning Ukraine, Gaza, Iran and Lebanon—reveals crucial operational distinctions.

In Ukraine, open-source data shows a tight integration of uncrewed platforms, software stacks, and infantry units into unified combat architectures, but does not yet offer confirmation of systematic deployment of fully autonomous lethal systems operating against human personnel entirely without human oversight.

Reporting by Reuters²⁷ notes Ukraine’s adoption of a new integrated drone-infantry combat system, while other sources indicate field-data sharing to train military AI models²⁸. Ukraine serves primarily as a testbed for accelerating automation²⁹, where reconnaissance, defensive postures, and target identification rely heavily on semi-autonomous tools and rapid data aggregation.

In Gaza, the deployment of algorithmic warfare is significantly more pervasive, largely because the occupying and genocidal force is the State of Israel—one of the world's most dominant powers in the surveillance and defense technology industries. In the Gaza Strip, AI-assisted tools have been deployed for drone interception, alongside autonomous UAVs engineered to navigate underground tunnel complexes. Meanwhile, investigative reports revealed the operational use of AI platforms such as Lavender and The Gospel (Habsora) to generate and prioritize target sets for large-scale bombardments. These examples are particularly significant for understanding the automated warfare of today and tomorrow.

According to investigations by +972 Magazine and Local Call, the Lavender system was designed to process population-scale data and flag individuals as potential targets based on suspected affiliation with the armed wings of Hamas or Palestinian Islamic Jihad. The algorithm was trained on data profiles of known militants to identify matching behavioral patterns across the broader civilian population—including communication networks, WhatsApp group memberships, frequent SIM card changes, and mobility anomalies.

The most critical revelation lies not in the existence of the software, but in how military operators processed its outputs. Sources cited in the investigation revealed that Lavender was deployed with minimal human oversight, effectively reducing human operators to "rubber stamps". Verification was often reduced to a cursory check lasting approximately 20 seconds per target—frequently limited to confirming whether the algorithm's selected target was male.

According to the investigation³⁰, in the initial weeks of the military campaign, the Israeli military reportedly relied almost entirely on Lavender, which flagged up to 37,000 Palestinians as suspected militants, linking their profiles directly to residential locations for potential airstrikes.

Journalistic investigations further revealed that Lavender operated alongside auxiliary tracking systems, including software colloquially known as "Where's Daddy?" It is therefore not merely a matter of identifying a suspect, but of automatically linking them to a private residence to create the ideal moment for an attack—ensuring strikes were conducted at night while entire households were present.

According to testimony gathered, the primary objective was to eliminate entire family units alongside the designated target. This information was obtained from sources within the Israeli Defense Forces (IDF), though official Israeli military spokespersons have broadly rejected the claims.

Conversely, The Gospel (Habsora) serves a different function³¹: rather than flagging specific individuals, it generates targets such as buildings, infrastructure, housing units, and operational sites. Described by sources as an AI-driven platform capable of generating targets "almost automatically" at a rate far exceeding human analytical capacity, it operates as a "mass assassination factory".

While these tools do not formally eliminate human authorization, they automate key nodes of the targeting pipeline—drastically compressing analytical timelines and scaling target generation to unprecedented levels.

In the Iranian and Lebanese theaters, algorithmic warfare continues to expand, though verification remains challenging. Dr. Gould frames these operations as a continuation of an operational arc—stretching from Afghanistan and Ukraine through Gaza—that progressively normalizes AI-enabled target acquisition. Reporting by The Wall Street Journal³² indicates that U.S. forces integrated Anthropic's Claude language model via defense platforms to support target identification workflows within command structures, processing over 1,000 targets during the initial 24-hour operational window.

Assessments from British think tank Chatham House³³ confirm U.S. Commander Brad Cooper's deployment of advanced AI suites to synthesize vast datasets during regional operations, accelerating intelligence workflows from hours to seconds³⁴. The think tank also cites the Maven Smart System³⁵ as an AI-enabled command-and-control platform designed to correlate multi-source intelligence, prioritize targets, and streamline operational planning, while noting it remains unclear how extensively it was deployed in Iran.

According to Palantir³⁶, the platform collects and integrates multi-source intelligence to provide commanders with an operational picture for identifying, classifying, and prioritizing targets.

Palantir traces the platform's origin back to Project Maven, launched by the U.S. Department of Defense in 2017 to integrate computer vision and machine learning into tactical analysis³⁷. In March 2025, NATO secured procurement agreements for the Maven Smart System, initiating a rapid rollout across all 32 member states within a record six-month timeframe³⁸.

According to official procurement statements from Ludwig Decamps, Director General of the NATO Communications and Information Agency (NCIA), the agency collaborated with Supreme Headquarters Allied Powers Europe (SHAPE) and Palantir to provide the alliance with cutting-edge, customized AI capabilities needed on the modern battlefield.

The statement notes that the platform allows commanders and service members to leverage next-generation AI safely in critical military operations, improving intelligence fusion, target identification, and battlefield awareness, while accelerating decision-making.

Taken together, these active theaters demonstrate that autonomous weapons must be analyzed within the broader ecosystem of remote and algorithmic warfare. While technical autonomy varies across regions, the operational trajectory remains uniform: the expansion of systems that categorize, rank, and recommend targets; the marginalization of human judgment; the proliferation of compounding civilian harm; and the erosion of clear legal accountability across distributed decision-making chains.

It is within this systemic gray zone—rather than the trope of the fully independent "killer robot"—that the most profound transformation of warfare is currently unfolding.

From Hawija to the United Nations

The trajectory toward algorithmically mediated remote warfare is rooted in the evolution of precision strike doctrine. Promised as a means to strike targets with heightened accuracy from greater distances—and with reduced political risk—precision warfare developed throughout the 20th century, prompting a technological, cultural, and doctrinal shift.

While the 1991 Gulf War is often cited as the formal combat debut of "precision warfare," scholars such as Dr. James Patton Rogers argue that its operational logic originated in World War I, when fire control, aerial reconnaissance, and artillery coordination first relied on systematic data collection.

In short, precision warfare was not defined by the guidance systems of smart missiles, but by the reliance on high-speed data processing to apply force. This trajectory accelerated through the development of uncrewed platforms, satellite surveillance, and integrated ISTAR (Intelligence, Surveillance, Target Acquisition, and Reconnaissance) networks, shifting military decision-making from an isolated human choice to a distributed technical process.

British sociologist Dr. Lucy Suchman demonstrates³⁹ how data automation platforms like Project Maven represent the logical extension of this precision doctrine: maximizing visual coverage, accelerating target classification, and executing strikes with greater discrimination.

However, historical precedents demonstrate the inherent fragility of "precision" narratives. On June 2, 2015, an airstrike conducted by two Dutch F-16s⁴⁰ targeted an ISIS munitions factory in Hawija, Iraq. Pre-strike intelligence assessed that no civilians were present in the immediate area, but the attack ignited massive secondary explosions from stored industrial explosives, leveling the surrounding residential neighborhood. The strike killed at least 85 civilians⁴¹ and injured hundreds more—representing one of the single deadliest civilian-harm incidents of the anti-ISIS campaign.

The Hawija incident undermines the premise that technical precision guarantees operational control: a precisely designated target can still produce catastrophic, indiscriminate effects. Furthermore, institutional investigations focused almost exclusively on the immediate blast radius, disregarding the severe long-term compounding harm—the loss of income for survivors, ongoing medical costs, displacement, and community destruction.

Those who bomb from afar tend to lose ground-level context. For years, the Netherlands failed to adequately acknowledge the extent of the damage, leaving affected populations without legal remedies, formal recognition, or compensation. Hawija stands as a critical case study of the opacity and diffusion of responsibility that occur when warfare becomes increasingly digitized and remote.

When applied to autonomous and AI-driven weapons, this history demands looking beyond narrow metrics of technical accuracy or formal legal compliance. Instead, it requires evaluating the broader ecosystem of systemic violence, invisibility, and diffused responsibility produced by the intersection of automation, big data, and remote operations.

Consequently, the international regulatory debate has expanded beyond fully autonomous kinetic platforms to encompass AI-based Decision Support Systems (DSS). UN General Assembly Resolution 79/239⁴² explicitly addresses artificial intelligence in the military domain, acknowledging that DSS platforms exert a profound influence over the identification, classification, and prioritization of targets—even when they do not directly trigger kinetic force.

This intermediate role makes DSS legally and ethically complex: these platforms constrain human cognitive agency, accelerate the kill chain, and redistribute legal responsibility across complex technical networks. While public debate often envisions "human-in-the-loop" oversight as an operator actively approving a strike, scholars like Dr. Lauren Gould emphasize that DSS shape outcomes long before kinetic authorization—for instance, during the identification and ranking phases where the set of options available to the human decision-maker is constructed. By filtering options, ranking priorities, and assigning algorithmic risk scores, the software constructs the operational reality within which the human operates; the human is not deciding from scratch, but within a pre-framed algorithmic pipeline.

Furthermore, the standard legal requirement of "meaningful human control" fails to define which human retains responsibility across the deployment pipeline: the system designer, the team selecting training data, the analyst setting risk thresholds, the commanding officer authorizing the attack, or the tactical operator given seconds to confirm an algorithmically prioritized target. Ultimately, human agency is distributed across a fragmented chain of actors with vastly different roles, technical capabilities, and operational authority.

The Rise of European Defense Tech

by Andrea Daniele Signorelli

Just a few years ago, the defense tech sector was largely shunned by investors, especially in Europe. Now, however, it stands as one of the technology sectors delivering the fastest growth rates. The reasons behind this turnaround are readily apparent: the proliferation of war zones and escalating geopolitical tensions have profoundly shifted attitudes toward companies and startups specializing in military technology.
Across Europe, these drivers are far more acute. Russia’s February 2022 invasion of Ukraine and Washington's threats under Donald Trump to abandon its allies—or drastically scale back its commitment—have convinced European NATO members, with the exception of Spain, to raise military spending to 5% of GDP by 2035. This target allocates 3.5% strictly to defense and 1.5% to military-grade civilian infrastructure. Meanwhile, the European Union has launched a rearmament program—known as Readiness 2030 (also referred to as ReArm Europe)⁴³—which aims to mobilize up to €800 billion by 2030, with the goal of “massively ramping up European industrial production for defense.”
Consequently, in 2025, European nations are projected to spend over €150 billion on the purchase of military equipment and weapons⁴⁴—more than double their 2021 spending, narrowly outpacing the United States. According to an analysis by the Bruegel think tank⁴⁵, between 67% and 90% of this capital—depending on the country—is absorbed by the ten largest Western defense manufacturers (in Europe, primarily BAE Systems, Thales, Leonardo, Airbus, and Rheinmetall AG)⁴⁶.
Yet this new arms race is enriching more than just traditional defense giants. Startups proliferating across Europe are rushing to supply rapid-turnaround hardware, munitions, and advanced tech solutions. “It is no longer possible to simply stick to the traditional schedules of months and years of planning, procurement, or development,” a spokesperson for DIANA (Defense Innovation Accelerator for the North Atlantic)—the NATO agency connecting defense startups with the Alliance’s armed forces—noted to Business Insider⁴⁷.

Who Is Funding European Defense Tech

As a result, since the invasion of Ukraine alone, more than 230 DSR (defense, security, and resilience) companies have emerged in Europe, securing a total of $8.7 billion in venture capital investments in 2025—55% higher than the previous year and fourfold the amount from 2021. By comparison, the broader European venture capital market has expanded by only 16% over the past 12 months. Artificial intelligence dominates, as expected: 44% of total funding flowed to companies leveraging AI to power drones and other autonomous systems, decision-support platforms, image recognition software, and more.
This tech-war race is funded primarily by public entities. The most significant player is the NATO Innovation Fund (NIF): launched in 2023 with a budget of €1 billion and backed by 24 allies, it stands as the world’s first “multi-sovereign” venture capital fund, dedicated to startups developing technologies vital to the security of the North Atlantic Alliance with an investment horizon of up to 15 years.
Complementing the NIF are instruments established by the European Union—such as the European Defense Fund (nearly €8 billion available for the 2021–27 window) and the new AGILE program⁴⁸—alongside initiatives rolled out by individual nations. France fields the public fund Definvest and the state-owned bank Bpifrance; Germany relies on innovation agency SPRIND⁴⁹; Estonia deploys a €100 million state fund via SmartCap; and Italy channels capital through CDP Venture Capital, which—while lacking a dedicated defense fund—allocates €1 billion to national security priorities like AI and cybersecurity.
Within the private sector, prominent players active in Europe include U.S.-based firms Lightspeed, Lux Capital, and General Catalyst—a titan currently managing over $43 billion in assets (with U.S. investors accounting for roughly 40–50% of capital flowing into European defense tech). Among leading European investors is Prima Materia, launched by Spotify co-founder Daniel Ek and Shakil Khan, which has pledged €1 billion to European “moonshot” startups (ambitious, high-risk projects). Other key European players include Klaus Hommels’ Lakestar,¹⁸ which has raised over €2 billion since its inception and directs an increasing share of its portfolio toward defense, and Plural, founded in 2022 with an initial €250 million fund followed by a second €400 million fund in 2024.
Beyond large generalist funds, a wave of specialized defense and security investors has expanded from just eight firms prior to 2022 to fifteen more launched between 2024 and 2025. These include D3, Expeditions, Darkstar, Green Flag Ventures, ScaleWolf, Hyperion, and Dutch firm Keen, backed by €180 million. A sharper indicator of the sector’s momentum is Germany’s DTCP, which in January 2026 launched Project Liberty⁵⁰, a €500 million fund dedicated exclusively to DSR startups—marking the largest private European venture capital initiative explicitly focused on the field.

Helsing, the European Defense Tech Leader

Capturing a major share of this capital inflow is German company Helsing, founded in 2021 by Torsten Reil, Gundbert Scherf, and Niklas Köhler. Ranked as the world’s second-largest defense tech startup by capital raised⁵¹ (trailing only U.S.-based Anduril), Helsing has secured €1.5 billion in investments from funds like Prima Materia, General Catalyst, Lightspeed, and Accel, driving its valuation to €15.5 billion (it also participates in the Land Tactical Collaborative Combat project, funded by the European Defense Fund).
Helsing’s flagship product is the HX-2 loitering munition—commonly known as a kamikaze drone—delivered monthly by the hundreds to Ukrainian and German forces. In February 2026, Helsing signed a €269 million contract with the German army, with potential expansion up to €1 billion⁵². Weighing roughly 12 kilograms with a 100-kilometer range, HX-2 drones navigate via stored maps, bypassing GPS reliance to resist electronic jamming.
HX-2 drones also coordinate in swarms using Altra, a decision-support and “recon-strike” (tactical precision strike) platform built by Helsing. Beyond drone coordination, Altra aggregates real-time telemetry from sensors, radar, electronic warfare systems, and ground intelligence. It utilizes computer vision to map the battlefield, detect threats, and—under human supervision—automatically assign them to appropriate weapon systems, ranging from artillery to drones. Operational with the Ukrainian military, Altra serves as a European counterpart to platforms like Palantir’s Maven.
Pipeline projects include the CA-1 autonomous fighter jet, slated for its maiden demonstration flight in 2027. It will run on Centaur⁵³ air combat software, which has already cleared flight trials on a Saab Gripen E fighter. Also in the aviation development pipeline is Cirra, an AI system engineered to detect and neutralize anti-aircraft radars, currently being integrated into the Eurofighter by its manufacturing consortium: Airbus, Leonardo, and BAE Systems.
Expanding into other warfare domains, Helsing has initiated production of the SG-1 Fathom underwater drone, designed primarily to protect subsea infrastructure (tested by the British Royal Navy in December 2025 under the “Atlantic Bastion”⁵⁴ program, focused on North Atlantic surveillance). In June 2026, the company unveiled the RX-1 quadruped robot—an experimental hardware-software platform for developing autonomous ground robotics capabilities, developed alongside institutions like ETH Zurich and INRIA Paris.

Other Key Players

Alongside peers like Tytan Technologies, ARX Robotics, and Quantum Systems, Helsing is headquartered in Munich. The Bavarian capital has solidified its status as Europe’s premier defense tech hub, capturing $1.7 billion of the $8.7 billion invested across the European sector in 2025. This dominance is by design: Munich offers a deep engineering talent pool from the Technische Universität München, close proximity to defense clients, and manufacturing muscle inherited from the automotive sector—an industry currently under pressure but readily repurposed to scale drone and weapon manufacturing.
Quantum Systems itself represents Europe’s second defense tech unicorn, boasting a valuation exceeding €3 billion and roughly €650 million in total funding from backers like DTCP, Balderton Capital, Hensoldt, Airbus Defence and Space, and the U.S. Department of Defense. Like Helsing, Quantum Systems equips its drones with proprietary AI software, pushing 2025 revenues past €300 million through contracts with the armed forces of Ukraine, Romania, Germany, and the United States⁵⁵. In another sign of tightening collaboration between startups and legacy defense contractors, Quantum Systems has forged strategic partnerships with Airbus Defence and Space for AI-powered aerial reconnaissance and with Hensoldt to integrate drones, sensors, and data management into broader defense architectures.
The third unicorn is Finnish firm ICEYE, valued at roughly €2.4 billion following a late 2025 funding round led by General Catalyst, with backing from Bpifrance and Finnish sovereign wealth fund Solidium. Unlike its peers, ICEYE does not build drones; instead, it operates the world’s largest constellation of synthetic aperture radar (SAR) satellites, delivering high-resolution, all-weather imaging day and night. Originally deployed to monitor Arctic ice, it exemplifies dual-use technology with civilian (disaster management, insurance) and military utility. Clients include the armed forces of Poland, Germany, Finland, Sweden, Portugal, and—since 2022—Ukraine, which utilizes its tactical overhead intelligence. In December 2025, ICEYE established a joint venture with German giant Rheinmetall to supply space reconnaissance data to the German military under a €1.7 billion contract.
Portuguese company Tekever is the most established player among the frontrunners: founded in 2001 and backed by the NATO Innovation Fund, its valuation exceeds €1 billion. Today, it manufactures large unmanned aircraft (its flagship AR5 features an 8-meter wingspan and weighs roughly 180 kilograms) built for long-range reconnaissance, surveillance, and target acquisition across land and open seas. Key customers include the Ukrainian military and the British Ministry of Defense.
In the era of drone warfare, aerial systems take center stage, yet other defense tech sectors are surging. Decision-support software represents the first major expansion area, aggregating field data into actionable operational scenarios. French firm Comand AI, founded in 2023, specializes in this domain: its platform analyzes environments and threats, proposes courses of action, and refines future missions using operational feedback. British firm Labrys Technologies operates along similar lines, building coordination and management systems for military, humanitarian, and crisis response teams.
Simulation and wargaming represent a second major growth vector. Leading this space is British company Hadean, which constructs virtual environments for troop training, combat simulation, and operational testing prior to field execution. The objective is to compress preparation timelines and mitigate risks by evaluating tactics in virtual sandboxes. Alongside partnerships with BAE Systems, Hadean secured a £20 million contract with the British Ministry of Defense in 2024.
London-based Portal Biotech, meanwhile, pioneers AI-driven protein sequencing technology designed to swiftly detect proteins, pathogens, and biological or synthetic hazards. Recognizing the critical nature of this field, the NATO Innovation Fund selected Portal for its inaugural investment in the sector.
In ground robotics, Estonian firm Milrem Robotics leads with its THeMIS unmanned vehicle—a tracked platform capable of cargo transport, casualty evacuation, reconnaissance, mine clearance, or weapon mounting. Deployed in active service in Ukraine, the company reports its systems are integrated into robotic programs or operated across 19 nations, including the U.S., UK, and Germany. A younger contender, Munich-based ARX Robotics, builds modular autonomous ground vehicles for reconnaissance, transport, tactical support, and training.
Low-cost air defense forms another vital frontier: the race to field affordable interceptors capable of neutralizing drones without exhausting multimillion-euro missiles like the Patriot. British firm Cambridge Aerospace, valued near $1 billion, has developed Skyhammer, a counter-drone interceptor priced at a fraction of conventional missiles—estimated in the tens of thousands of dollars. In April 2026, the British Ministry of Defense ordered hundreds of units⁵⁶, with initial deliveries rolling out in May and prospective exports to Persian Gulf allies. Estonian firm Frankenburg Technologies operates within this same sector.
Finally, AI-driven software is transforming weapon design. British company PhysicsX, founded by former Formula 1 engineers and nearing a $1 billion valuation, has engineered software capable of simulating material and component behavior in seconds, drastically cutting design cycles for aircraft and armaments. Its client roster includes major European defense conglomerates like Leonardo.
PhysicsX takes the core promise of defense tech to its logical conclusion: making military systems more cost-effective, rapidly prototyped, highly automated, and seamlessly integrated. Should these advanced projects fulfill their promise, they could substantially reinforce Europe’s abrupt push toward strategic military autonomy.

PopulAI: Simulating the Chaos of War

Led by Craig Beddis, London-based defense tech firm Hadean—established in 2015—builds “synthetic environments”: digital twins of the real world where military forces and governments can train, plan, and experiment. Its latest system debuted at the DSEI trade show in London in September 2025: dominAI, a command-and-control software suite that aggregates field data to inform command decisions. However, the segment that has driven Hadean’s partnerships with Palantir, BAE Systems, and NATO is its suite of synthetic environments and wargaming tools, most notably populAI.
Traditional military simulations map battlefields with static units, objectives, rules, and predictable outcomes. PopulAI, by contrast, populates simulation environments with the messy realities troops face during crises: civilians, traffic flows, information streams, critical infrastructure, and complex ambiguities.
While complex scenario preparation once demanded weeks or months of manual labor—yielding static models far removed from real-world friction—populAI leverages generative artificial intelligence to automate and dynamically update scenarios.
The platform generates real-time “emergent” behaviors (unprogrammed phenomena spawned by interacting variables), alongside hyper-local population data and regional contexts. The goal is to flood simulated environments with traffic, civilian populations, communications chatter, interference, and unexpected contingencies, forcing personnel to train not just against an adversary, but within a tangled socio-infrastructural ecosystem.
By integrating regional datasets, the platform constructs environments that behave chaotically and unpredictably. Trainers can modify scenarios on the fly without reprogramming, test alternative courses of action, review decisions retrospectively, and interlink third-party simulators within a unified workspace. This capability resolves a long-standing interoperability bottleneck in military simulation, where 2D maps, 3D engines, and training suites traditionally operate in silos.
Hadean’s software is already commercialized and operational. For the British Army, the company contributed to the ACTS (Army Collective Training Service) program, delivering a simulation demonstrator field-tested in Kenya⁵⁷. For NATO, it designed a shared digital environment enabling multinational joint operations simulation, plan testing, and AI-driven analysis. Furthermore, alongside defense prime BAE Systems, Hadean anchors Project OdySSEy, a multi-domain training environment originating from a 2022 agreement that went live in 2023.
In December 2024, the UK Ministry of Defense signed a framework agreement with Hadean valued at up to £20 million, underscoring London’s strategic commitment to the technology. In March 2026, a bridge funding round was led by Booz Allen Ventures⁵⁸—the corporate venture arm of the U.S. defense consultancy, marking its first transaction outside the U.S.—alongside the public British Business Bank and NATO-aligned Twin Track Ventures. Though valuation details remain private, the company positions itself as a prime candidate for British defense “unicorn” status.
Hadean exemplifies Europe’s push to build unified virtual environments where training, planning, mission testing, and command-and-control converge.

Hypersonica: The Goal—European Hypersonic Missiles

Hypersonica is a British-German defense startup founded in December 2023 by Philipp Kerth and Marc Ewenz, two Oxford PhD graduates incubated at TUM Venture Labs at the Technical University of Munich. Headquartered near Munich with a London subsidiary, the company aims to secure European sovereignty in hypersonic flight—defined as speeds exceeding Mach 5 (five times the speed of sound)—without relying on U.S. technology. The combination of extreme velocity, atmospheric skimming trajectories, and maneuverability renders hypersonic weapons exceptionally difficult to intercept.
While the missile lacks a public designation, a prototype underwent its first test flight in early February 2026 from Norway’s Andøya test range, reportedly topping Mach 6 (over 7,400 km/h) with a range pushing past 300 kilometers, with all avionics and control systems performing within expected parameters⁵⁹.
Crucial technical specifications—including mass, propulsion systems, terminal guidance, and launch platforms—remain classified. The reported range is currently a development target rather than a validated metric; the Andøya test utilized a demonstrator mounted on a partner rocket rather than an operational missile. The flight served primarily to validate navigation, avionics, control software, and sensor performance under extreme thermal and aerodynamic stress.
The most striking innovation lies in methodology: Hypersonica claims it can compress the leap from concept to production from years into months via modular architectures and automated simulation, undercutting traditional development timelines and astronomical costs (such as the U.S. Dark Eagle program, spanning decades with costs per missile nearing $40 million).
Historically, European hypersonic efforts focused exclusively on defense rather than offensive development. Hypersonica’s offensive pivot stems directly from the October 2024⁶⁰ Anglo-German Trinity House defense pact, followed by a May 2025 joint initiative to build strike systems exceeding a 2,000-kilometer range. The UK subsequently launched a seven-year, £1 billion national hypersonic initiative, with The Times reporting that Hypersonica was selected among 90 participating organizations in 2024⁶¹.
Initial seed funding totaled €2.5 million, led by U.S. fund General Catalyst. On February 10, 2026, the company secured a €23.3 million Series A round, led by Plural (which also backs Helsing) alongside General Catalyst, 201 Ventures, and Germany’s SPRIND innovation agency.
The roadmap targets an initial short-range strike capability in 2027 and a fully operational sovereign hypersonic deterrent by 2029. Significant hurdles remain, however: technically, aerodynamic friction above Mach 5 generates extreme heat, complicating guidance and stability under conditions that have challenged heavily funded state programs. Politically, deploying a weapon with a range exceeding 300 kilometers triggers stringent international arms control regulations, shortening warning windows and amplifying decision-making pressures—the political cost of securing independent strategic autonomy from Washington.

The New Leviathan: Weapons, Technology, and Raw Materials

by Davide Del Monte

Tracing the financial flows and networks that fuel the autonomous weapons industry requires decoding how once-distinct sectors have hybridized. The merger—or, more precisely, the hybridization—of the extractive industry, the technology sector, and military apparatuses has forged an integrated supply chain controlled by a handful of global players who consolidate decision-making power capable of reshaping geopolitical balances.

Journalists reporting on the darker corners of Silicon Valley frequently invoke⁶² the image of the Leviathan conjured by seventeenth-century philosopher Thomas Hobbes: the giant gripping a sword in one hand and a crozier in the other, symbolizing the merger of state and church into a single authoritarian power.

By contrast, the new Leviathan rising from San Francisco Bay roots its power in vast natural resources, massive data pools, and raw computing capacity. It operates as a sprawling techno-industrial colossus whose skeleton, muscles, and nervous system are built from extractive, military, and tech sectors, all coordinated and directed by a single central brain: finance.

This symbiotic fusion of military forces, tech monopolies, and capital cloaks an authoritarian threat in the rhetoric of individual and state security, projecting it into an unprecedented, insidious dimension. We are no longer dealing with public institutions bound, in theory, by voter accountability and legal oversight, but with a many-headed monster that delegates force, surveillance, and human life itself to opaque, unaccountable algorithms.

By automating state power and privatizing security infrastructure globally, this new Leviathan operates entirely outside institutional checks and balances. It threatens to reduce citizens from legal subjects to mere data points to be monitored, predicted, neutralized, and, if deemed necessary, eliminated.

Moreover, whether examining projects built by legacy, publicly traded multinationals or those incubated inside early-stage startups, the massive capital driving the Leviathan's expansion invariably traces back to the same financial architects. To maximize their dominance, these institutions strategically scatter their virtually limitless economic resources across every relevant sector.

A crucial distinction emerges between traditional industry—where established multinationals across mining, technology, and defense contribute distinct yet integrated parts to the production of autonomous weapons—and the rising wave of startups that vertically integrate multiple stages of the manufacturing pipeline within their own organizations.

These two alternative business models—which we can define as “traditional” and “innovative”—also reflect a sharp divergence in the entities deploying capital to coordinate development and production. While legacy development involves asset management goliaths and index-investing giants, the startup ecosystem is driven by aggressive venture capital funds.

Across both sectors, total command over supply chains and production processes allows daring investors to secure a commanding foothold at every step, from raw mineral extraction right down to the market launch of new weapon systems.

The Devil All the Time

In his 2020 film The Devil All the Time, director Antonio Campos transports audiences to Knockemstiff, Ohio, in the late 1950s and early 1960s. Life in this isolated town plays out with relentless monotony: faces change, but victims and perpetrators remain locked in an inherited cycle of violence from start to finish. Generations endure perpetual tragedy, while the structural distribution of wealth and power stays entirely unchanged.

Across the geography of contemporary global capitalism, corporate balance sheets bear an unsettling resemblance to that remote Ohio county. Whether looking at silicon producers in Japan, chip fabrication plants in California, or cloud infrastructure for the Pentagon, the exact same names recur at every annual shareholder meeting, casting votes to advance converging interests and strategies.

Reviewing the beneficial ownership disclosures⁶³ that U.S. publicly traded corporations must file with the SEC—the U.S. government agency responsible for overseeing and regulating financial markets—three institutions appear most consistently: BlackRock, Vanguard, and State Street. Familiar names inside and outside Wall Street, these giants firmly grip the levers of the war economy while maintaining distinct operational profiles.

Founded in New York in 1988 by Larry Fink, BlackRock recognized early on the immense potential of artificial intelligence, building its own proprietary investment management tool: Aladdin software. Functioning as the digital brain of global finance and used by governments, banks, and trading desks to track trillions of dollars, Aladdin propelled BlackRock to become the world’s largest asset manager, commanding an asset portfolio totaling approximately $14 trillion⁶⁴.

BlackRock’s tech ambitions extend far beyond its proprietary algorithm. Fink’s firm holds equity stakes ranging from 6% to 8% in virtually every major U.S. technology giant, including Nvidia, Apple, Microsoft, Amazon, Micron Technology, Intel, Meta, and Alphabet. Its footprint is equally formidable across the mining and defense sectors, where it holds comparable stakes in nearly every major player.

While BlackRock’s rise was fueled by software infrastructure and institutional ties to governments and central banks, The Vanguard Group grew out of a different, initially radical concept pioneered by founder John C. Bogle, encapsulated in his famous dictum: “Don’t look for the needle in the haystack; just buy the haystack!”

A champion of retail investors initially mocked on Wall Street for his unorthodox philosophy, Bogle founded Vanguard in Maryland in 1978, quickly transforming it into a mutual fund colossus. Unlike BlackRock, Vanguard manages approximately $10 trillion, is not publicly traded, and is owned directly by its constituent mutual funds, which are held by retail investors. Profits flow back to investors, driving management fees down drastically.

These structural differences, however, do not prevent close alignment with BlackRock. Bogle’s firm holds equity stakes of roughly 7% to 8% across nearly all major tech companies (including 9.63% of Apple), alongside positions exceeding 9% in major defense and mining firms—peaking at 11.9% in Albemarle Corporation, a global lithium mining leader.

The final member of this triumvirate, dubbed “The Big Three,” is State Street. Founded in Boston in 1792 as Union Bank and reshaped in the 1970s by William S. Edgerly into an investment powerhouse, State Street today manages approximately $5.6 trillion in assets⁶⁵. Famous for installing the iconic “Fearless Girl” statue in front of the Wall Street Bull in 2017, State Street holds significant stakes—exceeding 6%—in top U.S. defense contractors like Lockheed Martin, Northrop Grumman, and RTX Corp.

The ties binding these three financial giants go far beyond shared attendance at corporate shareholder meetings; their relationship runs much closer than one might imagine. Publicly traded since 1980, State Street counts BlackRock as a holder of 8.99% of its shares, while Vanguard held over 13% until an internal organizational restructuring in March 2026 reduced that stake to zero⁶⁶. Executive linkages tightened further in 2024 when Salim Ramji—who spent the prior ten years holding senior executive roles at BlackRock—was named CEO of Vanguard.

Academic circles frequently study another hallmark of their coordination: the systematic alignment of votes at corporate annual meetings, commonly termed the “corporate governance oligopoly.” As studies published by Harvard University⁶⁷ demonstrate, each of the three firms casts its proxy votes following guidelines that mirror the others, effectively standardizing corporate decisions on crucial issues like defense spending, the green transition, technological development, and dividend payouts.

While the Big Three are joined by other heavyweights like Fidelity Investments, Goldman Sachs, J.P. Morgan, and Capital Group, their dominance across the tech–raw materials–defense triad remains unmatched. To unpack how this power translates across the entire production cycle—from raw material extraction to finished weapon systems—we can examine two strategic U.S. defense programs: Lockheed Martin’s autonomous onboard system for the F-35 fighter jet and Project Linchpin, one of the Pentagon’s most sophisticated command-and-control platforms.

Automatic Target Recognition on the F-35 Lightning II

Lockheed Martin proudly describes its flagship product, the F-35 Lightning II, as “the world’s most lethal, resilient, and connected fighter jet,”⁶⁸ a status driven in part by specialized software and hardware systems capable of advanced autonomous operation.

Among these, one of the most compelling features of the fighter is the Automatic Target Recognition (ATR) system, an artificial intelligence capability that transforms the jet into an autonomous sensing weapon capable of identifying and isolating targets on land or at sea using computer vision neural networks.

Specifically, the software is engineered to process data from onboard sensors—such as infrared cameras, radar, and electro-optical systems—in real time with complete autonomy to identify, classify, and track stationary or moving targets.

To function, the ATR system relies on ultra-high-precision optical instruments like the Electro-Optical Targeting System (EOTS)⁶⁹, which require specialized magnets manufactured using rare earth elements, including neodymium and dysprosium.

These raw materials are supplied by MP Materials, a U.S. company and global leader in rare earth extraction with which the U.S. Department of Defense signed a $35 million agreement⁷⁰, to establish the nation's first heavy rare earth processing and separation plant, supporting both commercial and military applications.

This agreement is particularly revealing because it highlights Washington’s attempt to break its reliance on China⁷¹ for rare earth supplies by subsidizing American firms like MP Materials, whose shareholder roster includes The Vanguard Group with a 7.7% stake and BlackRock with a 6.6% stake.

In November 2024⁷², the Finnish company ICEYE joined the fray, leading a project to develop artificial intelligence technology capable of enhancing ATR performance under harsh weather and lighting conditions along NATO’s northern flank. This move underscores a clear strategic push to adapt cutting-edge U.S. military hardware for future theaters of global conflict, such as the Arctic. Just one month after the partnership was announced, BlackRock participated in the Finnish company’s funding round with a €65 million investment.

Running local vision algorithms directly onboard the aircraft in milliseconds—without relying on internet connectivity—demands immense computing power. L3Harris Technologies⁷³ handles this bottleneck via hardware like the Integrated Central Processor (ICP), the brain of the F-35 Lightning II fighter jet's ATR suite. Major institutional shareholders of L3Harris include BlackRock with 7.2%, The Vanguard Group with 12.2%, and Capital World Investors with 6%.

Furthermore, the software installed on fighter jets must be capable not only of "seeing" and processing surroundings in a fraction of a second, but also of accurately recognizing them. To remain effective, algorithms are pre-trained on the ground using millions of digital images and massive digital storage capacity, bringing Microsoft⁷⁴, Amazon⁷⁵, and Palantir⁷⁶ into the fold via military cloud contracts for JWCC cloud services⁷⁷ and fleet data analytics.

As Lockheed Martin notes in the press release announcing its Microsoft partnership, “the future of air defense will not be won by a single sensor, countermeasure system, or algorithm. It will be won by a unique ecosystem capable of detecting, optimizing, and mitigating.”

Investment fund exposure runs deep across this link in the development chain. The Vanguard Group holds stakes in Amazon and Microsoft, owning 7.2% and 8.95% of their shares, respectively, while BlackRock wields influence with a 5.9% stake in Amazon, 7.3% in Microsoft, and 6.9% in Palantir.

Tracing the path from raw mineral extraction all the way to the F-35's onboard ATR system reveals the footprints of The Vanguard Group and BlackRock at every step, culminating at prime contractor Lockheed Martin. On the shareholder roster of the Bethesda defense giant, the three major funds hold dominant stakes, led by State Street at 14.9%, followed by The Vanguard Group at 8.9% and BlackRock at 7.4%.

Project Linchpin

While Lockheed Martin’s F-35 fighter jets dominate public discourse due to their exorbitant cost, Project Linchpin—announced by the Pentagon in 2022⁷⁸—remains a low-profile initiative worth examining.

As documented on SAM.gov⁷⁹, the portal dedicated to U.S. military procurement, Project Linchpin is not a physical weapon, but rather a command, control, and intelligence (C4ISR) software architecture engineered to provide “reliable solutions for AI-based systems through secure artificial intelligence operations and services (AIOPs+), while maximizing the investments of the Department of Defense (DoD) and the intelligence community (IC).”

The infrastructure aggregates feeds from a multitude of battlefield sensors—drones, satellites, and radar—to build a secure environment in which to train, test, and continuously update artificial intelligence algorithms. The objective is to establish a continuous workflow for retraining models in the field so they adapt to evolving threat scenarios without requiring a complete system overhaul.

Powering the command-and-control operating system that trains thousands of neural networks in real time demands peak computing power across centralized servers and mobile edge nodes.

At this stage, familiar corporate entities re-emerge: Amazon⁸⁰ and Microsoft⁸¹ provide militarized cloud infrastructure under the Joint Warfighting Cloud Capability (JWCC) framework with IL6 (Top Secret) security clearance, operating alongside tech giants like Oracle⁸² and Nvidia, manufacturer of the high-performance Blackwell GPUs.

Both Amazon and Microsoft are heavily backed by BlackRock and The Vanguard Group, as is Nvidia, where the two funds hold 7.4% and 7.3% of shares, respectively. By contrast, a June 2025 SEC filing shows that only The Vanguard Group holds a stake in Oracle, standing at 5.3%.

Manufacturing Nvidia GPUs and connecting server racks requires ultra-high-purity silicon and copper, sourced primarily from Japanese leaders Shin-Etsu, SUMCO Corporation, and Ibiden, alongside Taiwan’s TSMC and U.S.-based Freeport-McMoRan.

Major U.S. investment funds maintain heavy equity positions across these raw material suppliers, starting with Freeport-McMoRan, a Phoenix-based mining giant and premier global producer of copper, gold, and molybdenum, in which BlackRock holds a stake exceeding 6%.

The financial triumvirate's reach extends across the Pacific into the shareholder base of Taiwan Semiconductor Manufacturing Company (TSMC)⁸³, alongside major U.S. investment firms like JPMorgan, Capital World, FMR, and Goldman Sachs.

In Japan, State Street and JPMorgan appear prominently among top shareholders in Shin-Etsu⁸⁴ and SUMCO⁸⁵. State Street also holds a stake of just over 2% in Ibiden⁸⁶, a world leader in high-tech electronic components and advanced ceramics.

The shadow of American finance looms over the entire pipeline, converting quartz, silicon, and copper into the costly, advanced chips without which Linchpin and other AI-based platforms could not function.

The U.S. military's ambitious Linchpin project incorporates several other contractors, including Red Hat (an IBM subsidiary providing foundational system software), Booz Allen Hamilton (specializing in AI and digital transformation), and innovative startups⁸⁷ like Enveil and Latent AI, which focus on encryption and model compression. Latent AI maintains close ties to both Booz Allen Hamilton and Lockheed Martin, both of which invested in the startup through their corporate venture capital arms in 2021.

These two case studies demonstrate the pervasive presence of the Big Three—BlackRock, The Vanguard Group, and State Street—at every single tier of the autonomous weapons supply chain. This near-absolute, highly concentrated corporate power spans every industrial sector, drawing intense blowback not only from researchers and activists, but also from political figures.

In 2024, a coalition of eleven Republican-led states spearheaded by Texas filed suit against the three financial giants, accusing them of wielding decision-making power within rival coal companies to force production cuts that artificially drove up energy prices for American consumers. In a quintessentially American paradox, the eleven Republican governors accused the Big Three of undermining the national economy by capitulating to progressive climate policies and activist demands.

The Rise of “Little Tech”

“Over the past decade, a new generation of tech billionaires has emerged, presenting themselves not only as entrepreneurs but as visionary messiahs, convinced that technology can transform the world.”

These words, used by CBS News reporter Sharyn Alfonsi⁸⁸ to introduce her interview with Anduril founder Palmer Luckey, aptly capture the recent economic and ideological shift reshaping the U.S. defense market.

While massive traditional conglomerates across the mining, defense, and technology sectors—closely intertwined through the nerve centers of global finance—continued business as usual, a smaller but equally formidable new Leviathan began emerging from San Francisco Bay.

This new entity takes industrial hybridization to an extreme, where roles are no longer segregated among various, closely related industrial sectors; instead, individual entities leverage their manufacturing capabilities to cover almost the entire supply chain on their own.

Here we find the hundreds of startups that have proliferated in recent years, capable of operating successfully across technology, defense, and extraction while attracting intense political interest and, above all, billions of dollars in investment.

This multifaceted universe of economic actors seeks to “move fast and break things,”⁸⁹ constituting what Marc Andreessen and Ben Horowitz term “Little Tech” in contrast to the dominant Big Tech industry.

Andreessen and Horowitz, key figures in the sector, have not only actively contributed through their investments to the exponential growth of the tech and defense startup sector, but have also worked to reposition these startups within the national political and economic ecosystem, paving the way for new entrepreneurs such as Luckey.

In their manifesto The Little Tech Agenda⁹⁰, published in July 2024 on the website of their venture capital firm a16z, the two investors explain their worldview bluntly: “Little Tech is the term we use for tech startups, as opposed to the giants of Big Tech. Until now, Little Tech has operated independently of politics. But, as the old Soviet joke goes: ‘You may not care about politics, but politics cares about you.’ We believe that bad government policies are now the number one threat to Little Tech. We believe that American technological supremacy—and the crucial role that Little Tech startups play in ensuring it—is a political issue of the highest order, on par with any other. The time has come to take a stand in favor of Little Tech.”

Beyond the obvious financial interests of professionals deploying billions into extremely high-risk companies, the two partners also explicitly state the ideological goal that must underpin investments in this sector: the protection of “American technological supremacy.”

Unlike figures such as Fink, Bogle, Edgerly, and other big names in traditional finance, these new venture investors and entrepreneurs are far more radical in both their market operations and their political expressions—often conservative and nationalist views that inevitably shape the sector's development.

The differences between traditional investors and venture capitalists go beyond the frantic pursuit of ideological and political recognition; they are structural and systemic.

First, while the large funds discussed earlier operate primarily in public markets by investing in publicly traded companies, venture capital funds operate exclusively in private markets, pooling capital from institutional investors or high-net-worth individuals to back companies deemed to carry exceptionally high potential alongside extreme risk.

Furthermore, while traditional funds exercise influence through proxy voting at shareholder meetings without participating in active management, venture capital firms acquire equity stakes in startups, frequently securing seats on boards of directors to ensure active participation in both strategic and operational decisions.

Finally, the most important distinction: the fundamental goal for venture capital funds is the “exit”—ideally within five to ten years—monetizing the investment through an initial public offering (IPO) or a buyout by a larger company.

This exit mandate drives the golden rule that, as Catherine Bracy explains in her essay World Eaters⁹¹, underlies the strategy of any ambitious venture capital firm: the power law (or power-law distribution).

In short, while most human events follow a normal distribution—where the average represents the norm and extremes are rare, such as human height or weight—venture capital is governed by a radically asymmetric curve: a very small number of investments (unicorns) must generate nearly all profits, while the vast majority of startups fail or barely recoup invested capital.

This dynamic, Bracy explains, leads venture funds not to seek out structurally sound companies capable of slow, steady growth, but rather companies with the mathematical potential to “pay back the entire fund” on their own if the investment hits the mark.

This aggressive approach forces funded startups to prioritize speed over operational efficiency in order to eliminate potential competitors from the field. Drawn from military jargon, this growth strategy—termed “blitzscaling”—often drives aggressive entrepreneurs to make radical moves, such as entering markets without regard for regulatory frameworks, required permits, or official approvals, as seen in the infamous cases of Uber and food delivery platforms.

Among the most active venture capital funds is Andreessen Horowitz’s a16z, which serves as a major conduit for high-risk, high-potential investments and—perhaps above all—a powerful engine for supporting U.S. military capabilities. A standout asset in a16z’s portfolio is Anduril, whose mission is to “transform the military capabilities of the United States and its allies through cutting-edge technologies,” anchored by its flagship product, Lattice OS, an AI-based command-and-control platform for military operations.

Browsing through the products showcased on the company’s website—founded in 2017 by Palmer Luckey and four partners, almost all alumni of Palantir—one finds prototypes of autonomous aircraft like Fury and Barracuda, unmanned fighter jets, small underwater drones, and dozens of other high-tech weapons.

Since its founding, Anduril has attracted bold investors eager to back the industry’s new unicorn, which promises to “accelerate complex kill chains”⁹² through its flagship software. Following its latest funding round, finalized by Andreessen Horowitz on May 13, 2026, with $5 billion invested, Anduril’s total funding raised in just nine years has exceeded $11 billion.

Among Anduril’s investors are other leading venture funds in the autonomous weapons sector, such as Founders Fund and General Catalyst, alongside the U.S. Department of Defense (which issued a check for over $13 million in January 2025) and BlackRock (which participated in an August 7, 2024 funding round raising a total of $1.5 billion).

Palmer Luckey’s business philosophy is remarkably clear and explicit, summed up by a remark he made in his early thirties during the aforementioned CBS interview: “We need to go from being the world’s police to being the world’s gun shop.” For those with billions to invest in high-tech weaponry, this is welcome news.

A16z’s portfolio spans software-driven defense firms like Air Space Intelligence (ASI)⁹³ and Castelion⁹⁴, the latter founded by former SpaceX employees to build “the systems America and its partners need to maintain a global technological advantage.” It also features pioneers like ZeroMark, a defense tech venture leveraging artificial intelligence, robotics, and firearms to redefine security and public safety with products that “significantly increase the speed, accuracy, and precision of weapons while reducing collateral damage.”

While the list extends much further, the profiles of these three firms—straddling tech and defense with millions in backing from Andreessen Horowitz—illustrate the fund’s deep stake in autonomous weapons.

Returning to Anduril’s investor roster, two other names stand out: General Catalyst and Founders Fund. Founded in 2000 in Cambridge, Massachusetts, General Catalyst has been led since 2021 by Hemant Taneja, an Indian-American billionaire who, unlike his a16z peers, positions himself as a champion of ethical investing. Author of essays such as Intended Consequences and The Transformation Principles, Taneja openly critiques Silicon Valley’s dominant dogma—the “move fast and break things” mantra and its doctrine of maximizing short-term profit at society's expense.

Yet despite these claims, trumpeted across dozens of self-congratulatory executive essays, General Catalyst not only finances major foundational players in large language models like Anthropic and Mistral AI, but holds prominent positions in key autonomous weapons firms. These include Castelion, the Finnish startup ICEYE (which signed a collaboration agreement with Lockheed Martin), and the German firm Helsing, which collaborates with several European governments—such as on the cutting-edge Franco-German-Spanish Future Combat Air System (FCAS) program⁹⁵—to modernize ground-based air defense, electronic warfare, and algorithmic surveillance.

Unlike Taneja, Peter Thiel—co-founder and prominent face of Founders Fund—disdains ethical branding, embracing his reputation as an industry hawk. Founded in San Francisco in 2005 by Thiel and other members of the “PayPal Mafia”⁹⁶ Founders Fund now manages over $11 billion across hundreds of startup investments.

Founders Fund has poured capital into frontier technologies, backing PsiQuantum⁹⁷ (aiming to build the world’s first quantum computer), Impulse Space⁹⁸ (accelerating humanity’s future in space), and EnduroSat (making space accessible through affordable satellite services).

In the autonomous weapons sector, alongside Anduril and Thiel’s own Palantir, Founders Fund backs two standout startups: Hadrian⁹⁹, which automates both weapon manufacturing and its workforce, and Nominal¹⁰⁰, which develops validation software for complex industrial systems and military hardware.

In this landscape—cast as dystopian fiction two decades ago, but representing a lucrative investment landscape today—KoBold Metals¹⁰¹ is a notable case in point. Founded in 2018 with the motto “predict everything,” the company raised over $920 million between 2022 and 2024 from investors including Andreessen Horowitz, Mary Meeker’s Bond Capital¹⁰², Sam Altman’s Apollo Project, and notably Breakthrough Energy Ventures (BEV)¹⁰³.

The involvement of this last fund underscores KoBold’s strategic weight. Breakthrough Energy, launched in 2016 by Bill Gates to accelerate the global green transition, boasts backing from institutional investors and billionaires including Jeff Bezos (Amazon), Jack Ma (Alibaba), Richard Branson (Virgin Group), and Michael Bloomberg (Bloomberg LP).

The massive attention and capital lavished on KoBold Metals by the world’s richest individuals boils down to a single goal: addressing critical mineral shortages using artificial intelligence. Rather than managing mines directly, KoBold uses AI to locate deposits of critical metals—copper, cobalt, nickel, and lithium—that elude traditional geological surveys. These metals are essential for electric vehicle batteries, energy infrastructure, and, ultimately, autonomous weaponry.

KoBold Metals perfectly embodies the hybridization noted earlier: by leveraging advanced AI systems built by its tech-titan investors, the firm secures the raw materials required to manufacture those very systems efficiently and securely. Having raised $1.2 billion since inception, KoBold has achieved a historic milestone with the discovery of Mingomba in Zambia, hailed as the highest-grade copper deposit found in twenty years.

This achievement transcends commerce, serving as a vital geopolitical asset that has drawn intense White House focus as Washington deploys global strategies to counter China’s African mining monopoly.

The guest list for Donald Trump's state visit to Beijing from May 13 to 15, 2026, reads like a cross-section of the corporate network mapped throughout this investigation. Among the sixteen executives accompanying the delegation were Larry Fink of BlackRock, Meta's Dina Powell McCormick, Apple's Tim Cook, and Elon Musk of SpaceX and xAI, alongside Boeing CEO Kelly Ortberg and GE Aerospace CEO H. Lawrence Culp—both critical suppliers for the F-35 program despite competing with Lockheed Martin.

This delegation of American capital and technological power fulfills the blueprint Peter Thiel laid out at the ultra-conservative NatCon forum¹⁰⁴ in Washington in 2019. Thiel argued bluntly that the tech sector must be recognized as vital to state defense, lest the nation fall to its geopolitical rivals.

Specifically: China.

The European Union’s Push: Public Investment and Research

By Laura Carrer

Over the past decade, the EU has moved to strengthen its strategic autonomy by mobilizing substantial resources for the defense industry. The primary instrument of this industrial policy is the European Defense Fund (EDF), which allocates €8 billion for the 2021–2027 budget period.
Within this sector, public universities have become essential partners in developing advanced defense and surveillance systems for European countries. EDF grants finance projects across the entire span of emerging technologies, covering up to 100% of basic research costs and 20% to 90% of development. To qualify for funding, calls for proposals require consortia comprising at least three entities based in three different EU member states—a structure designed to foster collaboration centered on major industrial players.

This dynamic is illustrated by Leonardo, Europe's top contractor by number of projects secured between 2022 and 2024 (37), totaling €172 million. Leonardo is followed by Germany’s Indra Sistemas SA¹⁰⁵ (30 projects, €94 million) and, far behind, another Italian firm, Elettronica Spa. Resources are directed toward highly specialized domains, including space surveillance, cybersecurity, artificial intelligence for decision support, and electronic systems for autonomous platforms¹⁰⁶.

Examining the rationale behind European defense funding is essential to understanding its goals and current trajectory. The EDF did not emerge overnight as a sudden reaction to Russia’s invasion of Ukraine in February 2022. Its origins date back at least to 2016–2017, when the European Commission and the European Council began articulating the need for a more structured defense R&D policy.

Following the EU Global Strategy and the joint EU-NATO declaration in July 2016¹⁰⁷, the Commission adopted the European Defense Action Plan¹⁰⁸ that November. It was framed not as an effort to build a "European army," but to lay the groundwork for deeper defense cooperation. The following month, the European Council urged the Commission to submit concrete proposals for the future fund by the first half of 2017.
The assessment driving this shift was stark: between 2005 and 2014, European defense spending dropped by 11%, while Russia doubled its spending and China increased its own by 150%. In this light, 2017 was viewed as the last chance to arrest the decline of Europe's industrial and technological base and keep the continent competitive in a mounting global arms race.

Prior to the EDF, two initial initiatives were launched as tests. The first was the Preparatory Action on Defense Research (PADR), running from 2017 to 2019 with a budget of roughly €90 million. The second was the European Defense Industrial Development Program (EDIDP), active from 2019 to 2020 with a €500 million budget. Their purpose went beyond launching initial projects; they served as stress tests for the mechanisms, stakeholders, and eligibility criteria of the eventual EDF.

Specifically, implementing pilot phases helped neutralize political resistance among member states and tested a co-funding model where EU funds leveraged much larger national investments. The Commission and member states resolved to direct European funding toward the most hazardous and capital-intensive phases of innovation—prototyping and testing—where many tech startups flounder in the "valley of death," unable to commercialize their products and turn a profit.

Consequently, the EDF did not build Europe’s defense research sector completely from scratch. Almost all recipient institutions had spent years participating in programs like PADR, EDIDP, Horizon 2020, and Horizon Europe, or partnering with agencies like the European Space Agency (ESA). Instead, the funding locked these institutions into an explicitly military-industrial framework, using research to integrate supply chains, harmonize standards, and boost the competitiveness of Europe's defense base.

To qualify, projects must be transnational and involve entities from multiple member states—a requirement designed partly to curb reliance on non-EU suppliers in sensitive fields like artificial intelligence, cyberspace, advanced sensors, quantum technology, and autonomous systems.

At the NATO Summit’s Defense Industry Forum in The Hague on June 24, 2025, European Commission President Ursula von der Leyen noted¹⁰⁹: "Since the start of the war [in Ukraine], investment in European defense tech startups has surged by 500%. Yet funding remains scarce for the critical bridge from startup to scale-up."

The Role of Public Universities (and Beyond)

Europe’s race for defense technology has made universities key players in a new funding era. Extensive scientific and technical expertise is required for the research and subsequent development of technologies capable of operating in the defense sector—expertise that companies lack in-house.

Public institutions and universities thus find themselves in industry-led consortia focused on concrete military capabilities. They operate within a framework far more industrial than traditional civilian programs. This has inevitably created structural tensions that universities are beginning to address, either independently or collectively.

Major research institutions and universities receiving EDF funding between 2022 and 2024 include the Netherlands Organization for Applied Scientific Research (TNO)¹¹⁰. Founded in 1932, TNO is managed by an Executive Board appointed by royal decree upon the recommendation of the Ministry of Defense and the Minister of Economic Affairs and Climate Policy.

Other major recipients are the Austrian Institute of Technology (AIT)¹¹¹, the largest non-university research institute in Austria and one of Europe's largest research centers—whose shares are held by the Republic of Austria (50.46% via the Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation, and Technology) and the Association for the Promotion of Research and Innovation (49.54%)—and the Norwegian Institute for Defense Research (FFI)¹¹², the primary body responsible for defense research in Norway, which provides scientific and technological advice to the Ministry of Defense and the Norwegian Armed Forces.

In the autumn of 2023, following the Israeli offensive on Gaza triggered by the October 7 attacks by Hamas militias, protests on European campuses spread rapidly across multiple countries, including Italy¹¹³, Spain, Belgium¹¹⁴, Ireland, the Netherlands¹¹⁵, Finland, Denmark, and Austria. Through occupations, encampments¹¹⁶, and resolutions, students have demanded transparency, divestment, and the severing of ties with Israeli universities and research centers.

These events are significant for this report because Israel is one of the leading producers of surveillance technologies and artificial intelligence systems—including those used in warfare—that have been field-tested on the Palestinian population. In Italy, the University of Turin¹¹⁷ suspended its participation in a specific Italy-Israel scientific cooperation program in 2024 while maintaining other agreements.

In Spain, CRUE, representing 76 universities, announced it would review and potentially suspend partnerships with Israeli institutions over compliance with international humanitarian law¹¹⁸. In a specific case involving the University of Granada¹¹⁹ in March 2025, the suspension of academic and scientific collaborations with Israel was also upheld by the courts. The court ruled that the university "acted within the scope of the autonomy granted by relevant legislation and its own bylaws, which clearly prioritize a 'culture of peace, nonviolence, and cooperation among peoples'," and that "the suspension of collaborations with Israeli institutions does not constitute a discriminatory act, but rather a proportionate measure aimed at ensuring compliance with international humanitarian law."

In Belgium, the University of Antwerp and the Université Libre de Bruxelles suspended certain collaborations with Israeli universities while leaving open the possibility of completing some research projects already underway, subject to ethical review. Trinity College Dublin stated in a May 8, 2024, press release that it had decided to divest from Israeli companies and to also review exchange programs with Israel following a prolonged student protest¹²⁰.

Overall, campus protests have not shattered European scientific cooperation with Israel. Instead, they have triggered selective suspensions, ethical reviews, withdrawals from specific grants, and intense pressure to restrict research ties.

A May 2026 report by the League of European Research Universities (LERU)—which includes prestigious institutions such as the University of Amsterdam, the University of Milan, and University College London—notes, "we are witnessing significant changes in society’s expectations of universities regarding defense research, accompanied by widely divergent views on the legitimacy of participating in such research"¹²¹.

The debate, the report continues, centers on "what universities should do (normative), what they are authorized to do (legal/governance), and how they should make decisions (procedural)."
A foundational issue identified by the LERU report as "an unresolved ambiguity" is the definition of dual-use. European Commission President Ursula von der Leyen has stated that promoting dual-use research is part of the European Union’s response to a "more threatening geopolitical context."

Although the concept refers to goods, products, or technologies (including research support services) originally designed for civilian purposes that can also be deployed in military contexts or diverted from their original intent, classifying research as strictly "civilian" or "military" has grown increasingly complex. This is because almost any field of knowledge can ultimately yield military applications or become relevant to national security without triggering standard military controls.

In this context, a 2021 regulation by the European Parliament and the Council of the European Union established a landmark regulatory framework in the dual-use sector, establishing a supranational regime for controlling exports, brokering, transit, and transfer. Exporting these goods outside the EU requires a license; in Italy, this is issued by the national authority UAMA (Armaments Materials Authorization Unit) within the Ministry of Foreign Affairs and International Cooperation.

Months later, in September 2021, the EU urged research institutions to implement internal compliance programs for scientific work involving those items. The EU has thus catalyzed the race toward technological innovation in the military sphere while placing the responsibility on companies—and especially universities—to ensure compliance with stringent research security and export regulations.

A second ambiguity lies in distinguishing between strategic partners and systemic rivals. China, for instance, is simultaneously an important research partner, an economic competitor, a strategic rival, and a security threat—a dynamic that also applies to countries like Russia and Turkey.

As noted earlier, following campus protests, Israel has increasingly come to be viewed as a country warranting closer scrutiny. The LERU report explains that the vocabulary emerging around such countries includes terms like "strategic competitors," "systemic rivals," and "countries of interest," yet notes that none of these terms provides clear guidance for research collaboration decisions.

Ultimately, there are no strictly "friendly" or "hostile" countries, as almost no country fits neatly into either category. Universities must assess the strategic implications, geopolitical risks, and potential dual-use impacts of every collaboration on a case-by-case basis, despite lacking innate expertise to do so.

Among the most problematic issues identified by the LERU report is the tension between independent researcher autonomy and institutional academic freedom. When a computer scientist working on machine learning argues that their research is fundamentally civilian despite obvious applications in autonomous weapons, institutional restrictions must rely on controversial judgments about unknowable end-uses or impose broad limitations that restrict vast areas of legitimate science. Both approaches remain open to debate.
This raises a critical question about the stance universities should take on defense research. Given that the line between civilian and military research is increasingly blurred, "maintaining moral distance through selective abstention loses its coherence," the researchers note.
The LERU report cites a case where a university accepts funding for artificial intelligence security research from "civilian" agencies while rejecting defense agency funding, questioning whether this maintains principled distance or represents a purely symbolic measure with little practical impact on weapons development.

Universities also face complex decision-making processes regarding defense research. Which actors within the university—considered in its entirety—actually decide how research resources and time are spent? The faculty, a specialized group, students, or other actors?

The LERU report illustrates a scenario applicable to any Italian university working with a company like Leonardo: if a university community votes to ban weapons research, does the ban extend to artificial intelligence algorithms with potential applications in autonomous weapons? Each university establishes its own boundaries to reduce this ambiguity.

Some Italian universities participating in recent EDF projects have since established oversight mechanisms. Currently, it remains impossible to verify whether these mechanisms can effectively shut down funded projects that breach internal ethical standards. Even so, it marks a tangible shift for an academic landscape where the defense industry has become increasingly normalized both economically and politically.

Guidelines adopted by the University of Bologna in March 2026 cite the 2021 EU dual-use regulation, noting: “Research and innovation are based on openness, international cooperation, and academic freedom. Nevertheless, the current geopolitical context, marked by growing tensions and a strategic interest in advanced knowledge and technologies, exposes the European research system to potential risks.”

In June 2025, the university also formed a dual-use and research security working group to ensure regulatory compliance and develop training on research security. More broadly, the goal is to promote a culture of awareness and individual responsibility across the entire university community, focusing particularly on faculty, researchers, and administrative and support staff involved in international projects or cutting-edge technologies.

The primary tool is a self-assessment requiring researchers to answer three fundamental questions before launching any international collaboration: Does the research subject fall within the category of dual-use goods (such as AI, biotechnology, or cryptography)? Who is the client, is the client based outside the EU, and who will hold the intellectual property rights to the results? Are the international partners trustworthy, and who is the end user?

For the defense sector, the university demands even greater rigor: identifying strategic objectives transparently and ensuring outcomes do not conflict with university values. Collaborations with military entities must observe “strict adherence to legal and ethical standards.”

In a statement to the author, the University of Bologna's press office emphasized that “researchers are made aware of and held accountable for conducting adequate due diligence on project partners, research objectives, and the possible uses and end users of the results, in order to identify and manage any potential risks.”

Furthermore, the university notes that the Academic Senate, during its September 23, 2025 meeting, “adopted a resolution regarding institutional collaborations with entities based in Israel, defining its scope and parameters. This decision reflects the university’s attention to concerns raised by the academic community, in compliance with national and international obligations and the principles of academic autonomy and freedom.”

The university secured EDF funding of just over €1 million for the ARCHYTAS project¹²². The European Defense Fund’s website describes how “AI can enable autonomous vehicles, the strategic analysis of large volumes of data, intelligence gathering, and surveillance through advanced computer vision, opening up a wide range of defense applications.”

In light of the March 2026 guidelines and the DUSE research security committee, the university confirms that “should any sensitive or risky issues arise, the university will take action in accordance with the methodologies defined within the DUSE framework, adopting, where necessary, appropriate and proportionate mitigation measures.”

The University of Trento, an ARCHYTAS partner, stated in a freedom-of-information response that the project aims to explore AI accelerators for civil protection—activities dedicated to safeguarding civilians during emergencies. The university adds that it is “drafting its own guidelines on dual-use technologies in accordance with provisions established by the Ministry of Universities and Research.”

“Regarding the outbreak of the Israeli-Palestinian conflict,” the FOI response states, “beginning in 2024, there have been motions from the Student Council and the Academic Senate.” On October 22, 2025, a press release¹²³ noted that the Academic Senate of the University of Trento, “in line with the positions taken by many other Italian institutions […], expresses full support and appreciation for the signs of a resumption of dialogue” following the agreement reached at the time between the US, Israel, and Hamas.

The press release also condemned Israeli authorities “for the extremely serious war crimes committed, which have led to well-founded accusations of genocide against the Palestinian population.” Additionally, the response confirmed the establishment of a “Permanent Laboratory on Peace and Conflicts” at the university.

Unlike other institutions, the Politecnico di Milano has neither drafted guidelines for defense research nor established dedicated working groups. This stance is striking given the Polytechnic’s deep involvement in European Defense Fund projects: six separate initiatives totaling nearly €5 million in funding between 2024 and 2025, most focused on studying and prototyping technologies with military applications.

One of these is ARCHYTAS, alongside the University of Bologna; another is EMISSARY¹²⁴. The EU portal describes EMISSARY as a project that “aims to reach a phase of prototyping, demonstration, and qualification for a broad set of developed technologies, reflecting the commitment to promoting European military capabilities in the field of space security and surveillance [...]”.

Leonardo was awarded €20 million, Telespazio €3 million, Thales Alenia Space Italia €2 million, and Elettronica €1.2 million. The Polytechnic University of Milan received €600,000.

The participation of Italy’s primary defense contractor, Leonardo, in EMISSARY forms just one part of a long-standing collaboration with the Politecnico. A report by Studenti Indipendenti—a student group founded in 2019—in collaboration with Polimi4Gaza, an association of students and professors opposing the Gaza conflict, notes that Leonardo “has secured exclusive ownership of the research results” generated at the university¹²⁵. This means “the technology and knowledge developed by the Polytechnic cannot be utilized or commercialized outside the university without explicit company permission”—not uncommon given that EU project outputs frequently remain classified.

On transparency, the report notes that student representatives had to sign non-disclosure agreements (NDAs) to review company contracts, under threat of penalty. Despite this, Studenti Indipendenti points out that “the data reviewed by representatives remains vague and unspecific.”

In 2024, Il Sole 24 Ore¹²⁶ reported that a new heavy tank, the Main Battle Tank (MBT), was set to enter service to replace the Italian Army’s Ariete tanks, which have been in service for over thirty years. “The tank has established itself as an important strategic weapon during the war in Ukraine, after the significance of ground-based weaponry had already been rediscovered during the Iraq War (2003–2011),” the article noted.

In November 2024, the EDF-funded FMBTech project ("Technologies for Existing and Future MBTs") officially launched, with the Politecnico di Milano acting as the research institution (with an estimated budget exceeding €630,000) alongside Thales Italia and surveillance firm Cy4gate. Cy4gate has positioned itself since 2020 as a unique operator in Italy: as an IrpiMedia investigation in 2021 revealed¹²⁷, “it is the first Italian company to combine strict cybersecurity with interception services for international police forces and all-source intelligence.”

Formed as a joint venture between Elettronica and Expert System—both specializing in electronic warfare systems like anti-drone tools, threat detection, and communications surveillance—Cy4gate has secured contracts worldwide, spanning the UAE, Saudi Arabia, Pakistan, Qatar, Central Asia, and Latin America, alongside NATO countries and EU projects. “Many of these countries,” IrpiMedia notes¹²⁸, “have already been involved in abuses of surveillance technologies in the past.”

The FMBTech project involves implementing several technological upgrades to tanks currently used in Europe and Italy. The main battle tank (MBT)—as project descriptions state—serves as “[...] the centerpiece of a digitized capabilities framework,” aiming to provide “[...] greater interoperability with unmanned systems and increased crew efficiency through artificial intelligence.”

Detailed information on the Politecnico’s EDF projects was difficult to obtain. “The available information is what has already been published on institutional websites,” the press office stated, adding that “in general terms, the University conducts its research activities in compliance with regulations and the principles of responsibility, integrity, and research ethics.”
Another major research institution, the Polytechnic University of Turin, faced intense scrutiny in May 2026¹²⁹ after professors issued an open letter to the Academic Senate demanding a total halt to military research on campus.

A year earlier, following protests, an agreement signed by the university’s Research Ethics Committee took effect, committing the institution and its partners to ensuring research results would not be applied in military contexts.

However, according to La Stampa¹³⁰, certain entities collaborating with Turin Polytechnic—including Italian ministries—refused to sign the clause, forcing a revision of the guidelines. “In addition to the defense of Italy,” the article reports, “military research for the defense of the European Union is now permitted. Furthermore, since public administrations (including the Ministry of Defense, for example) are presumed to comply with the Constitution, they may henceforth refrain from signing the clause.”

The university press office explains that “the update to the regulations confirms that agreements with public administrations are governed by procedures established for relationships between public entities,” operating “in full compliance with the Italian Constitution.” The amendment “is intended to clarify and make internal processes more transparent.”

One of two projects funded at Turin Polytechnic under the 2024 EDF call is FESPAN: “A consortium of 25 organizations, research institutes, universities, industrial partners, and SMEs from 10 European countries,” according to the EU portal¹³¹. FESPAN aims to enhance “future situational awareness capabilities of European defense at the tactical and operational levels, contributing to the development of tools to support tactical decision-making [...].” The project has received strong backing from EU defense ministers, securing several letters of support that cleared the path for its approval by the EU.

In October 2025, Sapienza University in Rome adopted guidelines¹³² to provide its academic community with a framework for managing dual-use research. The university also established a Research Ethics Committee and a support group to monitor research activities involving non-EU partners, particularly private contracts.

The university classifies certain countries as high-risk for research collaborations, including the Russian Federation—“due to increasing restrictions on civil liberties, internal repression of political dissent, military actions in Ukraine, and allegations of violations of international humanitarian law in conflict contexts”—and Israel—“due to the treatment of Palestinians in the occupied territories, military operations in Gaza, and violations of international humanitarian law in those contexts.”

Research involving mass surveillance, censorship, or military operations also faces strict scrutiny. Partnering with the University of Murcia, Sapienza participates in EU-GUARDIAN, an EDF project funded with €13.5 million between 2022 and 2025. The consortium developing “the most ambitious European project aimed at exploring operations in cyberspace within the context of the concept of mosaic warfare”¹³³ includes Spain's Indra Sistemas SA (over €3 million), Airbus Cybersecurity (€1.8 million), and Leonardo (over €2 million), alongside the University of Murcia (nearly €1.8 million) and Sapienza (€856,000).

The EU portal notes that EU-GUARDIAN “aims to create an artificial intelligence-based solution capable of managing and automating a large part of incident management and cyber defense processes,” rendering the system “capable of detecting, mitigating, and responding to security challenges in a semi-automatic or automatic manner [...].”

Although the University of Milan has not recently received EDF funding, it formed a working group in September 2025 “composed of 18 members, including faculty, administrative, and support staff,” with “the goal of addressing, in accordance with national and European regulations, the issues of dual-use and research security.”

The university added that “guidelines have been drafted but are still undergoing an internal verification, review, and testing phase,” scheduled for publication by the end of 2026. Additionally, “the University of Milan has expanded the tools to safeguard transparency and legality by approving, during the Senate meeting on May 12, an ‘integrity clause’ applied to agreements entered into by the university in other areas as well [...]”: a clause explicitly requiring parties to guarantee that agreement subject matter does not involve applying or transferring knowledge, technologies, or methodologies related to war or military purposes.

Globally, the growing entanglement of universities in military research has triggered similar shifts. A Nature article outlines how the German government is urging institutions to overturn decades-old policies limiting research to peaceful purposes¹³⁴. This shift carries profound weight given Germany’s post-war history concerning academic independence.

As the LERU report notes, “constitutional protection of intellectual freedom and provisions on institutional autonomy reflect deep skepticism about aligning universities with the state security apparatus.” Although over seventy institutions maintain voluntary “civil clauses” barring military research, political pressure has led several German universities to revise or formally abandon them. In 2022, following Russia’s invasion of Ukraine, Berlin allocated €100 billion to bolster the armed forces.

Regional approaches in Germany are also shifting. In July 2024, Bavaria enacted a law¹³⁵ mandating school and university cooperation with the German armed forces, drawing widespread criticism. Earlier legislative efforts¹³⁶ in Bavaria sought to scrap civil clauses entirely in favor of mandatory cooperation with the military. In North Rhine-Westphalia¹³⁷, legislation clarified as early as 2019 that a commitment to peaceful society does not conflict with military research.

In July 2025, De Morgen reported that several Flemish universities are easing restrictions on military research, adopting new ethical frameworks tailored to a changed geopolitical landscape. “We are not naive about global developments,” VUB Rector Jan Danckaert told De Morgen. “If research serves the rule of law and democracy, we are ready to take on that responsibility. But purely military research will not suddenly become the norm.”

Fonti

1. https://www.atlanticcouncil.org/blogs/new-atlanticist/autonomous-weapons-are-the-moral-choice/ ↑
2. https://disarmament.unoda.org/en/our-work/conventional-arms/convention-certain-conventional-weapons ↑
3. https://www.esd.whs.mil/portals/54/documents/dd/issuances/dodd/300009p.pdf ↑
4. https://www.files.ethz.ch/isn/188865/Ethical%20Autonomy%20Working%20Paper_021015_v02.pdf ↑
5. https://www.theguardian.com/technology/article/2024/jul/14/ais-oppenheimer-moment-autonomous-weapons-enter-the-battlefield ↑
6. https://www.stm.com.tr/en/kargu-autonomous-tactical-multi-rotor-attack-uav ↑
7. https://www.army.mil/article/291074/u_s_army_awards_enterprise_contract_for_it_commercial_solutions ↑
8. https://www.nytimes.com/2026/04/12/technology/china-russia-us-ai-weapons.html ↑
9. https://www.nytimes.com/2026/04/12/technology/china-russia-us-ai-weapons.html ↑
10. https://www.theguardian.com/technology/article/2024/jul/14/ais-oppenheimer-moment-autonomous-weapons-enter-the-battlefield ↑
11. https://www.wired.com/story/eric-schmidt-is-building-the-perfect-ai-war-fighting-machine/ ↑
12. La lettera, in realtà, è del 1939 e fu materialmente redatta dal fisico Leó Szilárd, che chiese a Einstein di firmarla per via della sua autorevolezza. ↑
13. https://www.washingtonpost.com/technology/2026/03/04/anthropic-ai-iran-campaign/ ↑
14. https://www.japcc.org/articles/how-large-language-models-are-transforming-modern-warfare/ ↑
15. A questo proposito si veda anche: https://blogs.icrc.org/law-and-policy/2024/09/24/transcending-weapon-systems-the-ethical-challenges-of-ai-in-military-decision-support-systems/ ↑
16. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646 ↑
17. https://www.pdcnet.org/pdc/bvdb.nsf/purchase26?openform&fp=philtoday&id=philtoday_2021_0065_0003_0549_0569 ↑
18. https://www.euronews.com/next/2026/02/27/ai-chatbots-chose-nuclear-escalation-in-95-of-simulated-war-games-study-finds ↑
19. https://www.independent.co.uk/news/world/americas/project-maven-ai-us-airstrike-iraq-anthropic-b2929138.html ↑
20. https://www.theguardian.com/technology/article/2024/jul/14/ais-oppenheimer-moment-autonomous-weapons-enter-the-battlefield ↑
21. https://global.oup.com/academic/product/the-double-black-box-9780197520901?cc=it&lang=en& ↑
22. https://www.researchgate.net/publication/336361772_Designing_Trustworthy_AI_A_Human-Machine_Teaming_Framework_to_Guide_Development ↑
23. https://www.al-monitor.com/originals/2024/02/israel-deploys-new-military-ai-gaza-war ↑
24. https://www.hrw.org/report/2025/04/28/a-hazard-to-human-rights/autonomous-weapons-systems-and-digital-decision-making ↑
25. https://intimacies-of-remote-warfare.nl/events/interdisciplinary-workshop-on-who-are-humans-in-military-ai/ ↑
26. https://archive.vn/vSPu0 ↑
27. https://www.reuters.com/business/aerospace-defense/ukraine-says-it-is-employing-new-integrated-drone-infantry-warfare-system-2026-04-15/ ↑
28. https://www.investing.com/news/world-news/ukraine-deploys-new-combat-model-commander-says-russianheld-areas-recaptured-4616830 ↑
29. https://www.stopkillerrobots.org/resource/statement-on-ethical-considerations-to-unga-meeting-on-autonomous-weapons-systems/ ↑
30. https://www.972mag.com/lavender-ai-israeli-army-gaza/ ↑
31. https://www.972mag.com/mass-assassination-factory-israel-calculated-bombing-gaza/ ↑
32. https://www.wsj.com/livecoverage/iran-strikes-2026/card/u-s-strikes-in-middle-east-use-anthropic-hours-after-trump-ban-ozNO0iClZpfpL7K7ElJ2 ↑
33. https://www.independent.co.uk/news/world/americas/us-politics/us-military-ai-iran-war-congress-oversight-b2936805.html ↑
34. https://www.chathamhouse.org/2026/03/iran-war-highlights-creeping-use-ai-warfare ↑
35. https://www.bbc.co.uk/news/articles/cdrm52g4pl2o ↑
36. https://assets.ctfassets.net/xrfr7uokpv1b/25muZs93DY5XOUBtuh68yn/b40d9784f69917219f972890ceede986/Maven_One_Pager.pdf ↑
37. https://blog.palantir.com/maven-smart-system-innovating-for-the-alliance-5ebc31709eea ↑
38. https://www.ncia.nato.int/newsroom/news/nato-acquires-aienabled-warfighting-system ↑
39. https://www.tandfonline.com/doi/full/10.1080/21624887.2020.1760587 ↑
40. https://paxforpeace.nl/publications/after-the-strike/ ↑
41. https://onlinelibrary.wiley.com/doi/10.1111/anti.70070 ↑
42. https://docs.un.org/en/A/RES/79/239 ↑
43. https://commission.europa.eu/topics/defence/future-european-defence_it ↑
44. https://www.economist.com/business/2025/10/29/europes-defence-firms-are-flying-now-for-the-hard-part ↑
45. https://www.bruegel.org/policy-brief/reforming-european-defence-procurement-boost-military-innovation-and-startups ↑
46. https://people.defensenews.com/top-100/ ↑
47. https://www.businessinsider.com/europe-defense-tech-startups-founders-vc-boom-funding-deployment-2026-2 ↑
48. https://defence-industry-space.ec.europa.eu/eu115-million-programme-agile-and-rapid-defence-innovation-agile-2026-03-26_en ↑
49. https://www.sprind.org/en/words/magazine/position-sprind-mil ↑
50. https://www.dtcp.capital/news-and-insights/detail/dtcp-launches-eur500-million-fund-for-defense-and-security-technologies-in-europe/ ↑
51. https://newmarketpitch.com/blogs/news/defense-tech-top-startups-fundraising ↑
52. https://resiliencemedia.co/germany-awards-stark-and-helsing-contracts-to-deliver-next-generation-strike-drones ↑
53. https://www.saab.com/newsroom/press-releases/2025/saab-achieves-ai-milestone-with-gripen-e ↑
54. https://ukdefencejournal.org.uk/royal-navy-will-trial-sg-1-drone-subs-under-atlantic-bastion/ ↑
55. https://www.reuters.com/business/german-drone-maker-quantum-nears-funding-round-that-could-triple-valuation-2025-10-16/ ↑
56. https://www.forcesnews.com/epic-fury/skyhammer-ready-pound-shaheds-if-needed-mod-sends-air-defence-system-mid-east ↑
57. https://hadean.com/blog/hadean-in-kenya-deploying-a-cloud-native-simulation-for-live-british-army-exercises-in-africa-part-1-2/ ↑
58. https://investors.boozallen.com/news-releases/news-release-details/booz-allen-and-hadean-modernize-ai-wargaming-western-allies ↑
59. https://www.hypersonica.com/en/news/hypersonica-successful-test/ ↑
60. https://www.gov.uk/government/news/joint-statement-on-uk-germany-trinity-house-agreement-progress-15-may-2025 ↑
61. https://archive.is/rlxmn ↑
62. Si veda ad esempio: https://www.cambridge.org/core/journals/canadian-journal-of-philosophy/article/algorithmic-leviathan-arbitrariness-fairness-and-opportunity-in-algorithmic-decisionmaking-systems/3AA0ECA77F8622488E9DB0834287215B o concetti molto simili espressi da studiosi e studiose come Kate Crawford in “Atlas of AI”, Frank Pasquale in “The Black Box Society” o ancora James Bridle in “The new Dark Age”. ↑
63. The definition of a Beneficial Owner is based not only on the economic ownership of shares, but primarily on voting power. The SEC requires the disclosure of anyone holding at least 5% of a company's capital or voting rights. ↑
64. https://www.privatebankerinternational.com/news/blackrock-q1-profit-2-12bn/ ↑
65. https://www.tradingview.com/news/stockstory:48b431246094b:0-state-street-nyse-stt-delivers-strong-q1-cy2026-numbers/ ↑
66. From State Street's Schedule 14A filing with the SEC: "The Vanguard Group filed a Schedule 13G/A on March 27, 2026 reporting, as of March 13, 2026, 0% beneficial ownership of State Street Corporation, amending a Schedule 13G/A previously filed with the SEC on July 29, 2025 reporting that, as of June 30, 2025, The Vanguard Group reported a shared voting power of 356,370 shares, sole dispositive power of 35,743,751 shares and shared dispositive power of 1,468,997 shares, representing beneficial ownership of 13.43% of State Street's outstanding common stock, as of record date March 25, 2026. In the March 27, 2026 Schedule 13G/A filing, The Vanguard Group reported an internal realignment effective January 12, 2026 wherein certain subsidiaries or business divisions of subsidiaries of The Vanguard Group will report beneficial ownership separately from The Vanguard Group. As of April 2, 2026, no subsidiaries or business divisions of subsidiaries of The Vanguard Group have reported beneficial ownership of State Street under Section 13(d) and Section 13(g) of the Exchange Act." ↑
67. https://corpgov.law.harvard.edu/2022/12/13/big-three-power-and-why-it-matters/ ↑
68. https://www.f35.com/f35/index.html?_gl=1*fiwops*_gcl_au*MTYxMzk2NDMyNi4xNzc5Mjg3Mjc3 ↑
69. https://www.lockheedmartin.com/en-us/products/f-35-lightning-ii-eots.html ↑
70. https://www.war.gov/News/Releases/Release/Article/2941793/dod-awards-35-million-to-mp-materials-to-build-us-heavy-rare-earth-separation-c/ ↑
71. https://rareearthexchanges.com/news/rare-earths-and-the-f-35-untangling-fact-speculation-and-spin/ ↑
72. https://www.iceye.com/newsroom/press-releases/iceye-to-lead-industry-consortium-to-deliver-advanced-space-technologies-and-analytics-to-finnish-government-f-35-industrial-participation-program ↑
73. https://www.l3harris.com/all-capabilities/high-performance-integrated-core-processor-icp ↑
74. https://www.lockheedmartin.com/en-us/news/features/2025/defending-the-skies-lockheed-martin-and-microsoft-collaborate-on-next-genc-uas-technologies.html ↑
75. https://aws.amazon.com/it/solutions/case-studies/lockheed-martin-drives-innovation-with-aws/ ↑
76. https://investors.palantir.com/news-details/2022/Palantir-Announces-Collaboration-with-Lockheed-Martin-Focused-on-Accelerating-Aegis-and-Future-Integrated-Combat-System-Software-Delivery-Powered-by-Palantir-Apollo/ ↑
77. https://www.war.gov/News/Releases/Release/Article/3239378/department-of-defense-announces-joint-warfighting-cloud-capability-procurement/ ↑
78. https://breakingdefense.com/2024/08/army-plans-for-faster-ai-adoption-for-defense-too/ ↑
79. https://sam.gov/opp/2bf40843e09f4e119824f1d062d42b27/view ↑
80. https://aws.amazon.com/it/blogs/publicsector/defense-agencies-can-access-aws-secret-region-for-il6-workloads/ ↑
81. https://learn.microsoft.com/en-us/azure/compliance/offerings/offering-dod-il6 ↑
82. https://www.oracle.com/it/defense-intelligence/jwcc/ ↑
83. https://www.holdingschannel.com/institutional/holders-of-tsm/ ↑
84. https://www.shinetsu.co.jp/en/ir/stock-info/#hldr ↑
85. https://www.sumcosi.com/english/ir/share/shareholders.html ↑
86. https://www.ibiden.com/ir/stock/stock-basick-data/ ↑
87. https://www.army.mil/article/280162/accelerating_the_armys_ai_strategy ↑
88. https://www.cbsnews.com/news/palmer-luckey-ai-powered-autonomous-weapons-future-of-warfare-60-minutes-transcript/ ↑
89. “Move fast and break things” è il celebre motto coniato da Mark Zuckerberg, fondatore di Meta. ↑
90. https://a16z.com/the-little-tech-agenda/ ↑
91. “World Eaters, how venture capital is cannibalizing the economy”, di Catherine Bracy (Dutton, 2025) ↑
92. https://www.anduril.com/lattice/command-and-control ↑
93. https://www.airspace-intelligence.com/ ↑
94. https://www.castelion.com/ ↑
95. https://helsing.ai/newsroom/helsing-commissioned-for-ai-backbone-in-fcas ↑
96. The so-called PayPal Mafia refers to a group of former employees and founders of PayPal who went on to found or develop other major technology companies based in Silicon Valley. ↑
97. https://www.psiquantum.com/ ↑
98. https://foundersfund.com/company/impulse/ ↑
99. https://www.hadrian.co/#what-we-build ↑
100. https://nominal.io/ ↑
101. https://koboldmetals.com/about-us/ ↑
102. https://www.globaldatacenterhub.com/p/mary-meekers-ai-manifesto-why-the ↑
103. https://www.breakthroughenergy.org/people ↑
104. https://nationalconservatism.org/natcon-dc-2019/presenters/peter-thiel/ ↑
105. https://www.indragroup.com/en ↑
106. https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/777936/EPRS_BRI(2025)777936_EN.pdf ↑
107. https://www.clingendael.org/sites/default/files/2021-02/Alert_European_Defence_Fund_February_2021.pdf ↑
108. https://eur-lex.europa.eu/EN/legal-content/summary/european-defence-action-plan.html ↑
109. https://sciencebusiness.net/news/dual-use/von-der-leyen-reaffirms-dual-use-rd-plans-ahead-eu-budget-proposal ↑
110. https://www.tno.nl/en/about-tno/organisation/ ↑
111. https://www.ait.ac.at/en/ ↑
112. https://www.ffi.no/en ↑
113. https://www.youtube.com/watch?v=n5lyi_ul_lY&t=2s ↑
114. https://www.eunews.it/en/2024/05/29/university-of-antwerp-and-universite-libre-de-bruxelles-suspend-collaborations-with-israeli-institutes/ ↑
115. https://www.timesofisrael.com/ashamed-amsterdam-alumni-to-return-diplomas-after-school-cuts-ties-with-hebrew-u/#:~:text=decision%20to%20unilaterally%20cut%20ties ↑
116. https://www.youtube.com/watch?v=kw5maeqW7Vs&t=1s ↑
117. https://www.timesofisrael.com/italys-university-of-turin-to-exit-israel-research-pact-amid-gaza-protests/#:~:text=suspending%20a%20collaboration%20agreement%20with%20Israeli%20universities ↑
118. https://english.wafa.ps/Pages/Details/143945#:~:text=suspend%20collaboration%20agreements%20with%20Israeli%20universities ↑
119. https://www.bricup.org.uk/news-2/court-upholds-university-of-granada-decision-to-break-ties-with-israel/#:~:text=after%20numerous%20mobilizations%20promoted%20by%20students ↑
120. https://www.tcd.ie/news_events/articles/2024/student-encampment-and-blockade-at-trinity-to-end/#:~:text=Trinity%20will%20complete%20a%20divestment%20from%20investments ↑
121. https://www.leru.org/files/LERU-Position-Paper-Defence-Research-in-European-Universities-A-Conceptual-Framework_2026-05-26-115144_ahga.pdf ↑
122. https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/opportunities/projects-details/44181033/101167870 ↑
123. https://pressroom.unitn.it/comunicato-stampa/dalla-parte-della-pace-e-del-dialogo ↑
124. https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/opportunities/projects-details/44181033/101168237 ↑
125. https://altreconomia.it/ricerca-militare-al-politecnico-di-milano-il-caso-di-leonardo/ ↑
126. https://www.ilsole24ore.com/art/cos-e-main-battle-tank-nuovo-carro-armato-dell-esercito-italiano-e-chi-potrebbe-costruirlo-AGmC9HZ ↑
127. https://irpimedia.irpi.eu/sorveglianze-cy4gate/ ↑
128. Ibid. ↑
129. https://www.lastampa.it/torino/2026/05/08/news/politecnico_lettera_aperta_ricerca_militare_docenti-15614015/ ↑
130. https://www.lastampa.it/torino/2026/05/08/news/politecnico_lettera_aperta_ricerca_militare_docenti-15614015/ ↑
131. https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/opportunities/projects-details/44181033/101167641 ↑
132. https://drive.google.com/drive/u/0/folders/1jrNBqRCDTokQ-h_MBOiWFGsluLk65aSV ↑
133. https://cyber.ee/research/projects/eu-guardian/ ↑
134. https://archive.vn/Rv0ei#selection-1049.1-1061.326 ↑
135. https://www.spiegel.de/panorama/bildung/bayern-schulen-und-unis-muessen-mit-bundeswehr-kooperieren-a-26435573-6968-4bcb-b96a-c09ebeb05606 ↑
136. https://www.universityworldnews.com/post.php?story=20240423203348462 ↑
137. https://www.universityworldnews.com/post.php?story=20190716140835343 ↑