All articles
AI & Security19 min read

The Privatisation of Military Decision-Making: Artificial Intelligence, Private Technology Companies and the 2026 Iran War

AI-enabled systems such as Palantir’s Maven Smart System and Anthropic’s Claude supported intelligence analysis and targeting in the 2026 US campaign against Iran. This article argues that the privatisation of warfare now extends into the technological infrastructure through which states decide how force is used.

The Privatisation of Military Decision-Making: Artificial Intelligence, Private Technology Companies and the 2026 Iran War

Artificial Intelligence, Private Technology Companies and the 2026 Iran War

Zapryan Valkov PhD Researcher in International Relations · University of National and World Economy (UNWE), Sofia November 2026

Abstract

The United States military campaign against Iran, launched on 28 February 2026, represents an important development in the integration of artificial intelligence (AI) into contemporary warfare. Reports concerning Operation Epic Fury indicate that AI-enabled systems, particularly Palantir Technologies’ Maven Smart System and Anthropic’s Claude large language model, supported intelligence analysis, target identification, operational planning and battle damage assessment. Although human commanders formally retained authority over lethal decisions, the increasing reliance on privately developed algorithms raises questions about the changing distribution of influence within military decision-making.

This article examines the 2026 Iran campaign as a case study in the evolving privatisation of warfare. Using publicly available military statements, investigative reporting, technical analysis and international humanitarian law scholarship, it distinguishes confirmed AI involvement from claims of autonomous targeting that remain unsubstantiated. It argues that the privatisation of warfare is expanding beyond the outsourcing of military personnel and logistical services into the technological infrastructure through which states interpret intelligence, formulate operational choices and exercise military power. This development creates new forms of strategic dependency, complicates accountability and challenges conventional understandings of state control over the use of force.

Keywords: Artificial Intelligence; Privatisation of Warfare; Military Decision-Making; Private Technology Companies; Project Maven; Operation Epic Fury; Iran; International Humanitarian Law; Human Oversight; Military Sovereignty.

1. Introduction

On 28 February 2026, the United States launched Operation Epic Fury against Iran, initiating a large-scale military campaign alongside Israeli operations. The offensive involved advanced combat aircraft, cruise missiles, drones and extensive intelligence capabilities. Yet one of its most consequential features was not a new weapon but the reported integration of artificial intelligence into military planning and targeting. According to the Center for Strategic and International Studies (CSIS), the Maven Smart System supported US military operations involving more than 1,000 targets during the first 24 hours of the war, demonstrating the increased operational tempo associated with AI-enabled targeting infrastructure (Mande & Allen, 2026). The Washington Post subsequently reported that Maven, incorporating Anthropic’s Claude technology, assisted with identifying potential targets, generating coordinates and prioritising objectives (The Washington Post, 2026).

These developments raise a question extending beyond the immediate conflict: what happens to state authority over military force when private technology companies develop the systems through which operational decisions are increasingly informed? Historically, discussions of the privatisation of warfare have concentrated on private military and security companies (PMSCs), including organisations providing armed protection, training, logistics and combat-related services. The activities of contractors during the wars in Iraq and Afghanistan demonstrated how military functions could be delegated to commercial actors without formally transferring state command authority.

The integration of commercial AI introduces another dimension to this relationship. Companies no longer need to deploy armed personnel to influence military operations, as their technologies may shape how information is interpreted, which potential targets receive attention and how military choices are presented to commanders. This article therefore examines the extent to which the integration of privately developed artificial intelligence into military targeting systems transforms state control over military decision-making. Its central argument is that AI-enabled warfare is creating a form of technological privatisation of military decision-making: a process through which commercial systems increasingly mediate operational analysis and recommendations while states retain formal legal authority over the use of force.

The analysis employs a qualitative case-study approach, drawing on official military information, investigative journalism, independent technical assessments and international humanitarian law scholarship. These sources are evaluated according to their evidentiary value, with particular attention to distinguishing publicly confirmed operational practices from technological capabilities and unverified claims. Given the classified nature of military targeting procedures, conclusions concerning the precise contribution of individual AI systems must remain qualified.

2. Historical Context: From Private Military Contractors to Private Military Algorithms

The relationship between states and private actors in warfare is historically well established. Mercenary forces, commercial suppliers and military entrepreneurs have supported governments for centuries. However, the expansion of private military and security companies after the Cold War introduced new questions concerning sovereignty, accountability and the division of military responsibilities. The conflicts in Iraq and Afghanistan demonstrated the growing operational significance of contractors providing security, transport, intelligence assistance, equipment maintenance and logistical support. As the International Committee of the Red Cross (ICRC) has observed, the activities of PMSCs can include functions traditionally associated with regular armed forces, creating important legal and operational challenges (ICRC, 2013).

The 2008 Montreux Document addressed this development by reaffirming existing international legal obligations and identifying good practices relevant to states employing PMSCs. Although not legally binding in itself, the document established an important framework for examining private participation in armed conflict (ICRC & Swiss Federal Department of Foreign Affairs, 2008). Nevertheless, commercial participation in warfare has evolved beyond conventional military contracting. Modern armed forces increasingly depend on private companies for satellite communications, cloud computing, cybersecurity, intelligence software, data analysis and artificial intelligence. These technologies may operate far from the battlefield, yet their outputs can influence decisions taken within operational headquarters.

This distinction is analytically important. A private military company supplying armed personnel performs a relatively identifiable military function, whereas a technology company supplying an AI-enabled targeting platform may influence the information and recommendations upon which military action depends. The latter arrangement does not necessarily constitute the delegation of command. Rather, it represents a transfer of substantial elements of the analytical infrastructure supporting command into privately developed technological systems. This development can be understood as a movement from outsourcing military capabilities towards outsourcing parts of the processes through which military choices are formulated.

3. Operation Epic Fury: Artificial Intelligence in the 2026 Iran Campaign

The United States Central Command (CENTCOM) confirmed that Operation Epic Fury began on 28 February 2026. Its official fact sheet dated 23 March reported more than 9,000 targets struck during the campaign by that point, alongside extensive operations involving bombers, fighter aircraft, naval assets and unmanned systems (CENTCOM, 2026). The scale of the campaign provides the operational context for understanding AI-enabled military systems. Contemporary warfare generates enormous quantities of information through satellites, reconnaissance aircraft, drones, communications intercepts and intelligence databases. Integrating this information quickly enough to support military operations creates a significant analytical burden, which artificial intelligence can help address by identifying patterns, classifying imagery, correlating information and producing recommendations for human consideration.

In March 2026, Admiral Brad Cooper, commander of CENTCOM, publicly confirmed that US forces were employing advanced AI tools to process large volumes of information and accelerate decision-making. He also explicitly stated that human personnel retained final authority over decisions concerning whether and when to engage targets (Al Jazeera, 2026). This distinction is fundamental because AI participation in the targeting process is not equivalent to independent authority to employ lethal force. Reuters reported on 1 March that the Pentagon had used Anthropic AI services during the operation, although the precise applications were not initially disclosed (Reuters, 2026a). Subsequent reporting by The Washington Post provided additional information about Maven and Claude, while CSIS examined the wider technological architecture.

Taken together, these sources establish that AI-enabled decision-support technologies played a substantial role in the campaign. They do not, however, establish that AI independently executed most attacks, that every target originated from an algorithmic recommendation or that commanders automatically approved machine-generated proposals. The available evidence therefore supports characterising Operation Epic Fury as a major example of AI-assisted warfare rather than fully autonomous warfare. This distinction is essential because the strategic significance of AI does not depend solely on whether machines can independently employ weapons, but also on their capacity to influence the decisions preceding the use of force.

4. Maven Smart System and Anthropic’s Claude: The Technological Architecture

Project Maven originated in 2017 as a US Department of Defense initiative intended to accelerate the military application of artificial intelligence, initially with particular emphasis on analysing imagery collected by surveillance systems. Following Google’s withdrawal from the project in 2018 amid employee opposition to military AI development, Palantir became a central commercial partner in the programme (Mande & Allen, 2026). The Maven Smart System subsequently developed into a broader intelligence and decision-support platform integrating information from multiple military sources.

According to CSIS, Maven combines intelligence collection, visualisation, automated detection, target management and operational planning within a common software environment. Rather than requiring analysts to examine separate databases and software applications, the platform can consolidate information into an integrated operational picture. Computer-vision systems can detect and classify objects within surveillance imagery, while other software components can correlate information across datasets, associate detected objects with geographical locations and support assessment of potential military objectives. Large language models add another layer of functionality by enabling natural-language interaction with information, the summarisation of complex documents and the generation of analytical recommendations.

In November 2024, Palantir announced its partnership with Anthropic to make Claude available through Palantir’s Artificial Intelligence Platform in classified government environments. Claude could therefore support natural-language interaction with data, document analysis and aspects of operational reasoning within the wider software architecture. CSIS identifies several potential contributions of language models to Maven, including the interpretation of complex information, identification of patterns across numerous detections and generation of possible courses of action (Mande & Allen, 2026).

Investigative reporting by WIRED offers further insight into these capabilities. Its examination of Palantir demonstrations showed how AI-assisted interfaces could analyse battlefield information, propose operational options and assist military users in selecting available resources (Haskins, 2026). However, an important methodological distinction must be maintained: software demonstrations establish technological capabilities, not necessarily their precise operational use in Iran. Similarly, although Claude was reportedly integrated into Maven-related workflows, publicly available sources do not reveal which individual targets, recommendations or strikes depended on its output. The operational architecture can therefore be described with reasonable confidence, while the contribution of particular models to specific attacks remains uncertain.

5. From Intelligence Analysis to Target Selection

One of the most important consequences of AI integration is the acceleration of the military targeting process. Conventionally, targeting involves several interconnected activities: identifying potential military objectives, collecting and validating intelligence, assessing operational priorities, selecting appropriate means of engagement, obtaining required authorisations and evaluating the effects of attacks. AI-enabled platforms can assist with multiple stages of this process by identifying objects of military interest from satellite or drone imagery, correlating information from different sources and generating recommendations concerning priorities, available resources or possible courses of action. Following an engagement, they may also assist with analysing post-strike information to assess whether an intended objective was achieved.

The Washington Post reported that Maven and Claude helped generate and prioritise potential targets during the Iran campaign. CSIS subsequently assessed that Maven had considerably increased the speed and scale of targeting operations (The Washington Post, 2026; Mande & Allen, 2026). The strategic implications are substantial. When the time required to assemble and analyse intelligence decreases, armed forces may act against a larger number of potential targets within a shorter period. This can be advantageous when confronting mobile missile systems, rapidly changing battlefield conditions or adversaries capable of dispersing military assets.

Nevertheless, greater speed does not necessarily guarantee greater accuracy. An algorithm may process information rapidly while relying on incomplete, outdated or incorrectly classified data. Furthermore, the apparent precision of machine-generated coordinates may conceal uncertainty about the underlying intelligence. A distinction must therefore be made between identifying an object and determining whether attacking it is lawful. Recognising a building, vehicle or military installation is primarily a technical and intelligence task, whereas establishing that a particular object constitutes a lawful military objective under the circumstances requires contextual evaluation and the application of international humanitarian law.

Consequently, targeting efficiency should not be measured exclusively through the number of proposed or completed strikes. A system capable of generating recommendations at unprecedented speed must also support rigorous verification and meaningful human scrutiny. Otherwise, improvements in operational tempo may create additional challenges for accountability and the protection of civilians.

6. Human Oversight and the Problem of Algorithmic Authority

The United States has emphasised that human commanders continue to control decisions involving the use of lethal force. However, the formal retention of command authority does not automatically resolve concerns regarding algorithmic influence. In AI-assisted military systems, the relationship between human operators and technological recommendations can become increasingly complex. Military personnel may rely on outputs generated by software they cannot fully inspect or independently reproduce, while operational urgency may limit opportunities to challenge the assumptions underlying algorithmic recommendations.

This introduces the risk of automation bias: the tendency to place excessive confidence in machine-generated assessments. The ICRC has warned that military AI decision-support systems may encourage overreliance on automated outputs and amplify errors when users accept recommendations without sufficient independent verification (ICRC, 2026a). Dorsey (2025) similarly argues that AI-enabled targeting systems risk transforming contextual legal and ethical assessments into increasingly quantified forms of reasoning. Complex military judgements cannot necessarily be reduced to numerical indicators or algorithmic classifications, particularly when they require consideration of civilian status, proportionality and uncertainty.

The problem becomes particularly significant when recommendations appear authoritative because they are generated through sophisticated technological systems. A human commander may formally approve a target while remaining heavily dependent on the software’s prior classification, prioritisation and presentation of the available evidence. The relevant analytical question is therefore not simply whether a human authorised the strike, but whether that person possessed sufficient information, time, expertise and practical freedom to exercise independent judgement.

Meaningful human oversight requires more than a human signature at the end of an automated decision-making process. It requires the ability to understand uncertainty, question recommendations, request additional verification and reject proposed action. This distinction is essential when evaluating claims that AI-supported warfare remains under human control, particularly in military environments where operational speed is considered a decisive advantage.

7. The Minab School Attack: Civilian Harm and the Limits of Technological Precision

The attack on Shajareh Tayyebeh Primary School in Minab on 28 February 2026 represents one of the most serious civilian-harm incidents associated with the opening phase of the Iran campaign. Human Rights Watch examined satellite imagery, video evidence and the physical characteristics of the strike. Its investigation identified damage consistent with precision-guided munitions striking multiple structures in a compound associated with Iran’s Islamic Revolutionary Guard Corps, including the school, which was physically separated from other facilities (Human Rights Watch, 2026). Reuters subsequently reported that investigators were examining whether outdated intelligence information had contributed to the school’s inclusion in targeting data (Reuters, 2026b).

In September 2026, a United Nations fact-finding mission reported reasonable grounds to believe that the US attack on the school amounted to the war crime of making a civilian object the object of an attack. The mission reported 157 people killed, including 123 children, according to credible sources (United Nations Human Rights Council, 2026). These findings raise serious questions regarding target verification, intelligence quality and compliance with international humanitarian law. However, a crucial limitation remains: publicly available evidence has not established that Maven, Claude or any particular AI model selected the school as a target or caused the targeting failure. It would therefore be methodologically incorrect to present Minab as proven evidence of an AI-generated targeting error.

Nevertheless, the incident illustrates the importance of examining the relationship between information quality, operational speed and human verification. If a targeting process relies on outdated intelligence, accelerating that process through artificial intelligence will not necessarily correct the underlying problem. Indeed, automation may reproduce or amplify existing errors unless systems incorporate effective safeguards and reliable mechanisms for updating information. The Minab case consequently highlights a broader issue: technological sophistication cannot substitute for the obligation to verify military objectives and protect civilians.

8. International Humanitarian Law and Responsibility

International humanitarian law applies to military operations regardless of whether decisions are supported by artificial intelligence. The foundational principles of distinction, proportionality and precautions in attack remain applicable. Distinction requires parties to an armed conflict to differentiate between civilians and combatants, and between civilian objects and military objectives. Proportionality prohibits attacks expected to cause incidental civilian harm excessive in relation to the anticipated concrete and direct military advantage. Precautions require feasible steps to verify targets and avoid or minimise civilian harm. These obligations are not displaced by the use of advanced software.

The ICRC has emphasised that AI decision-support systems may assist with intelligence analysis and certain operational assessments, but they cannot replace the contextual human judgement required to comply with humanitarian law (ICRC, 2026a; 2026b). The integration of commercial AI nevertheless complicates questions of responsibility because several actors may contribute to the production of an operational recommendation, including the state contracting authority, intelligence personnel, software integrators, model developers, military planners and commanders. Their involvement does not create identical legal obligations or levels of responsibility.

The state and its armed forces remain subject to applicable international law. Individual criminal responsibility depends on the relevant legal elements and evidence concerning a person’s conduct and mental state, while corporate responsibilities may arise through domestic law, contractual arrangements or other applicable legal frameworks. It would therefore be inaccurate to assume that the use of privately developed AI automatically transfers legal responsibility for a strike to the technology provider. Equally, treating AI merely as a neutral instrument may overlook the influence of its design, training data, operating parameters and deployment conditions. Effective accountability requires investigation not only of the final authorisation but also of the information and technical processes that contributed to the decision.

9. The Privatisation of Military Decision-Making

The central significance of the 2026 Iran campaign lies in the evolving relationship between states and private technology companies. Palantir and Anthropic are not conventional private military companies. Their primary contribution is not the deployment of armed personnel or the direct command of military units. Instead, they develop technologies that support intelligence interpretation, operational planning and the generation of military recommendations. This article therefore proposes distinguishing three dimensions of contemporary military privatisation: operational privatisation, concerning the outsourcing of military functions such as security, logistics, training and equipment support; technological privatisation, concerning dependence on privately developed infrastructure, including software, communications networks and AI systems; and decision-support privatisation, concerning the incorporation of commercial technologies into processes that shape the military options presented to state decision-makers.

These dimensions overlap but should not be treated as identical. Decision-support privatisation is especially significant because it concerns the informational foundations of military authority. A private company may have no formal right to authorise an attack while nevertheless developing the system that identifies potential targets, structures operational information and recommends courses of action. This raises questions about the practical distribution of influence within state military institutions and the extent to which commercial technological dependencies affect operational autonomy.

The dispute between the Pentagon and Anthropic during 2026 illustrates these tensions. Anthropic sought to preserve restrictions concerning the use of its technology for mass domestic surveillance and fully autonomous lethal weapons, while the US government sought broader discretion over lawful applications. The subsequent conflict over contractual conditions demonstrated the strategic consequences of depending on privately controlled AI models within sensitive military systems (Mande & Allen, 2026; Haskins, 2026).

From a sovereignty perspective, dependence on a proprietary technological platform introduces risks involving system access, commercial negotiations, model replacement, cybersecurity and operational continuity. A government may retain exclusive legal authority over military force while lacking complete technological autonomy over the systems supporting its decisions. The extent of this dependency varies according to contractual arrangements, technical architecture and the availability of alternative suppliers. Nevertheless, the Iran campaign suggests that privatisation should increasingly be examined not only through the physical delivery of military services, but also through commercial influence over military information and decision-making processes.

10. Strategic Implications and Future Research

The integration of commercial AI into military operations presents several competing strategic effects. On one hand, AI-enabled systems can improve the capacity of armed forces to analyse intelligence, coordinate military resources and respond to rapidly changing operational conditions. On the other hand, faster targeting processes may increase operational pressure, reduce opportunities for verification and encourage dependence on algorithmic outputs. There is also a growing issue of technological concentration: when critical military capabilities depend on a limited number of commercial suppliers, contractual disputes or technological disruption can become matters of national security.

For international relations research, these developments suggest a need to revisit the relationship between sovereignty, military power and private enterprise. Traditional accounts of military privatisation often examine whether states delegate the performance of military tasks. AI-enabled warfare introduces a further question: to what extent do states depend on private systems for the formulation and evaluation of military choices? This question is particularly relevant when technological suppliers possess proprietary knowledge concerning algorithms, data-processing methods and software architecture that may not be fully accessible to military personnel.

Future research should therefore investigate whether AI integration changes the practical balance between human judgement and algorithmic recommendations within military targeting. Comparative studies could also assess how different states regulate commercial suppliers whose technologies directly support operational decision-making, including contractual provisions governing transparency, auditing and continuity of access. International legal scholarship should additionally examine whether existing accountability mechanisms provide sufficient oversight when multiple commercial technologies contribute to military action. These questions are especially important as AI-enabled systems become increasingly integrated into command-and-control infrastructures and national defence strategies.

11. Conclusion

The 2026 US military campaign against Iran offers an important case study in the technological transformation of warfare. Available evidence establishes substantial reliance on AI-enabled systems in intelligence analysis, target development and operational decision support. However, it does not demonstrate that artificial intelligence independently executed most military strikes or exercised autonomous command over the campaign. This distinction is essential for accurate academic analysis because the principal transformation is not necessarily the removal of humans from military decision-making, but the growing influence of privately developed systems over the information, recommendations and operational choices available to them.

Companies such as Palantir and Anthropic occupy a different position from traditional private military contractors. Their technologies can influence military operations without their personnel directly participating in combat. The resulting challenge extends beyond questions of technological capability and concerns sovereignty, accountability, human judgement and the relationship between commercial innovation and state military authority. The Minab school attack underscores the consequences of failures in military targeting and the continuing importance of intelligence verification and civilian protection. Although no public evidence establishes that AI caused that particular attack, the incident demonstrates why technological acceleration must not be confused with legal or operational reliability.

Ultimately, the 2026 Iran campaign suggests that the privatisation of warfare is entering a new phase. The central question is no longer only who performs military functions on behalf of the state, but increasingly who develops, controls and shapes the technological systems through which the state decides how military force should be used. Understanding this transformation requires moving beyond conventional distinctions between state armed forces and private military contractors towards a broader examination of how commercial technological capabilities are embedded within the exercise of military power.

References

  1. Al Jazeera. (2026, March 11). US military confirms use of ‘advanced AI tools’ in war against Iran. aljazeera.com
  2. Dorsey, J. (2025). The erosion of human(e) judgement in targeting? Quantification logics, AI-enabled decision support systems and proportionality assessments in IHL. International Review of the Red Cross, 107(930), 1041–1071. doi.org
  3. Haskins, C. (2026, March 13). Palantir demos show how the military could use AI chatbots to generate war plans. WIRED. wired.com
  4. Human Rights Watch. (2026, March 7). US/Israel: Investigate Iran school attack as a war crime. hrw.org
  5. International Committee of the Red Cross. (2013). Contemporary challenges to IHL – Privatization of war: Overview. icrc.org
  6. International Committee of the Red Cross. (2026a). Frequently asked questions: AI decision-support systems in armed conflict. icrc.org
  7. International Committee of the Red Cross. (2026b, November 6). Frequently asked questions: Artificial intelligence (AI) in the military domain. icrc.org
  8. International Committee of the Red Cross & Swiss Federal Department of Foreign Affairs. (2008). The Montreux Document on pertinent international legal obligations and good practices for states related to operations of private military and security companies during armed conflict. icrc.org
  9. Mande, M., & Allen, G. C. (2026, June 2). What is Maven Smart System, and what does it do? Center for Strategic and International Studies. csis.org
  10. Reuters. (2026a, March 1). US uses Anthropic AI, B-2 bombers and suicide drones in Iran strikes. reuters.com
  11. Reuters. (2026b, March 11). US may have struck Iranian girls’ school after using outdated targeting data, sources say. reuters.com
  12. The Washington Post. (2026, March 4). Anthropic’s AI tool Claude central to U.S. campaign in Iran, amid a bitter feud. washingtonpost.com
  13. United Nations Human Rights Council. (2026, September). Findings of the Independent International Fact-Finding Mission on the Islamic Republic of Iran, presented at the Human Rights Council’s sixty-third session. un.org
  14. United States Central Command. (2026, March 23). Operation Epic Fury: Fact sheet. centcom.mil

Research note

This article is based on publicly available information as of 7 November 2026. Details of individual targeting decisions, classified military procedures and the precise contribution of specific AI models remain subject to evidentiary limitations. The conceptual framework and interpretations presented here are the author’s analytical propositions rather than established judicial or official findings.