The Unmanned Security State: Autonomous Defense, Algorithmic Enforcement, and the Institutional Politics of Order

Full report

On this page

1. Technical Taxonomy of Unmanned and Autonomous Architectures

The integration of artificial intelligence into state defense and security platforms has transformed the operational dynamics of coercive power1. To analyze this structural shift, clear technical distinctions must be established between tiers of platform autonomy1. Misunderstandings surrounding remotely piloted vehicles, autonomous navigation, AI-assisted decision-support tools, and fully autonomous weapon systems obscure the distinct legal, ethical, and operational risks associated with each operational layer1.

Tier I: Remotely Piloted Systems (Human-in-the-Loop)

Remotely piloted platforms represent a direct teleoperated extension of human agency1. In this operational framework, a human operator maintains real-time control over flight trajectories, sensor positioning, and effector deployment via encrypted radio-frequency or satellite links1. Systems such as medium-altitude long-endurance (MALE) unmanned aerial vehicles (UAVs) and teleoperated ground platforms rely on low-latency command channels5. The human operator processes visual intelligence, interprets contextual surroundings, applies moral and legal discretion, and manually authorizes engagement commands1. The primary vulnerability of Tier I architectures is their dependence on continuous communication channels, leaving them susceptible to radio-frequency jamming, electronic warfare, and signal latency in contested electromagnetic operational environments6.

Tier II: Autonomous Navigation Platforms (Human-on-the-Loop)

Autonomous navigation decouples platform movement from continuous human piloting1. Vehicles operating at this tier utilize onboard sensor packages—including micro-electro-mechanical systems (MEMS), inertial navigation systems (INS), optical flow sensors, lidar, and simultaneous localization and mapping (SLAM) algorithms—to execute spatial trajectories, navigate terrain obstacles, and maintain flight stability in GPS-denied or heavily jammed environments7. Shield AI’s Hivemind autonomy stack exemplifies this capability, enabling uncrewed aerial assets to conduct tactical maneuvering through unmapped structures without reliance on external communication or satellite positioning7. Under Tier II architectures, autonomy is restricted to spatial locomotion; the software manages vehicle movement while a human supervisor monitors system health, retaining sole authority over operational objectives and payload deployment1.

Tier III: AI-Assisted Intelligence and Targeting Support (Human-in-the-Command-Loop)

AI-assisted intelligence platforms transition algorithmic processing from physical navigation to multi-domain data fusion and cognitive target identification2. These architectures ingest unstructured streams from diverse sensors—signals intelligence (SIGINT), full-motion video (FMV), synthetic aperture radar (SAR), thermal imaging, and open-source intelligence—processing them through deep neural networks to detect, classify, track, and prioritize potential targets5. Command environments such as Palantir’s Artificial Intelligence Platform (AIP) and Gotham, alongside project Maven Smart System derivatives, sit at this operational layer2. They synthesize disparate inputs into a single operational graphic, generating automated targeting recommendations for human officers2. Although the software correlates track data and calculates engagement priorities, legal authority remains with human commanders, who must evaluate and validate machine-generated recommendations before executing a strike1.

Tier IV: Autonomous Weapon Systems (Human-out-of-the-Loop)

Fully autonomous weapon systems (AWS) represent the threshold where software algorithms independently perform the critical functions of target selection, tracking, and engagement without subsequent human intervention4. Once activated by an operator, an AWS utilizes onboard computer vision, thermal imaging, or anti-radiation seekers to search a bounded spatial-temporal area, match sensor inputs against target parameters, and initiate kinetic force4. Systems operating near or within this tier include anti-radiation loitering munitions (such as the Harpy series) and counter-UAS interceptors like Anduril’s Roadrunner-M and Bolt-M, which execute terminal target intercept trajectories once initial launch parameters are met5. In these architectures, the operational latency of modern warfare compresses decision cycles beyond human cognitive capacity, effectively shifting agency from real-time human discretion to pre-deployment algorithm parameters2.

2. Real-World Capability Matrix: Enterprise Software Layers and Tactical Effectors

The contemporary autonomous defense domain is defined by the integration of enterprise software stacks and modular hardware effectors2. Defense technology providers build software ecosystems that act as digital nervous systems, linking distributed sensors and kinetic platforms into unified networks2. At the top of this software architecture sit enterprise data integration and strategic decision-support platforms2. Palantir’s AIP and Gotham integrate unstructured operational data across secure cloud environments (such as DoD Impact Level 5 for controlled unclassified information and Impact Level 6 for classified networks)10. These platforms process multi-domain inputs to provide predictive logistics, intelligence correlation, and automated targeting workflows for theater commanders2. Beneath the enterprise strategic layer, edge operating systems manage real-time tactical command and sensor coordination8. Anduril’s Lattice OS functions as a software-defined command and control engine that fuses raw telemetry from ground radar, optical turrets, airborne reconnaissance drones, and acoustic sensors into a unified 3D operational graphic8. By executing computer vision and machine learning models on edge processing hardware, Lattice filters out ambient environmental noise, correlates target tracks, and presents operators with real-time threat priorities5. Platform-level flight execution is driven by tactical autonomy suites such as Shield AI’s Hivemind7. Utilizing reinforcement learning and adaptive control theory, Hivemind enables uncrewed aircraft to execute complex maneuvers, formation flying, and tactical search missions in GPS-denied and electronically jammed environments without active pilot communication links7. When embedded into uncrewed combat platforms, Hivemind allows distributed swarms to share target positions and adjust tactical behaviors dynamically7. These software layers orchestrate a range of physical sensing platforms and kinetic effectors:

• Sentry Towers: Solar-powered surveillance towers deployed on 33-foot portable masts or 80-foot extended structures5. Equipped with 3D AESA microdoppler radar, infrared turrets, and edge processing, Sentry towers autonomously detect, track, and classify human and vehicular movements at ranges up to 7.5 miles, feeding track data directly into Lattice OS5.
• Ghost and Ghost-X UAVs: Expeditionary vertical takeoff and landing (VTOL) reconnaissance drones featuring quiet acoustic signatures, 80-minute flight endurances, and operational ranges of 25 kilometers11. Ghost-X processes onboard visual data using mobile compute units, providing tactical surveillance in contested environments11.
• Altius Platforms: Tube-launched autonomous air vehicles (Altius 600/700) capable of being deployed from ground vehicles, aircraft, or larger uncrewed systems to conduct intelligence, surveillance, reconnaissance, electronic warfare, or loitering strike missions11.
• Roadrunner and Roadrunner-M: Reusable VTOL autonomous air vehicles powered by twin turbojet engines8. Operating at high subsonic speeds, the Roadrunner-M variant carries a high-explosive warhead designed for rapid air defense intercepts8. If an intercept command is aborted, Roadrunner autonomously returns to base, lands vertically, and undergoes rapid refurbishment for near-zero-cost reuse8.
• Fury (FQ-44A): An uncrewed Collaborative Combat Aircraft (CCA) built for air dominance and strike roles8. Powered by Lattice OS and integrated with tactical autonomy software, Fury operates alongside crewed fighter aircraft at high subsonic speeds5.
• Bolt and Bolt-M: Backpackable, loitering strike munitions designed for tactical edge engagement11. Bolt automates flight maneuvers, allowing field operators to designate a visual target track and select an attack trajectory, after which the onboard software executes the terminal dive autonomously11.

3. Normative, Legal, and Institutional Frameworks

The deployment of automated sensing and target engagement systems presents significant challenges to existing international legal frameworks4. Central to this debate is how International Humanitarian Law (IHL)—specifically the core rules of distinction, proportionality, and precaution—applies to software-driven combat systems4.

International Humanitarian Law and the ICRC Framework

The International Committee of the Red Cross (ICRC) maintains that IHL applies fully to all weapon systems, regardless of their level of autonomy4. However, the ICRC stresses that compliance with IHL requires human evaluation of context, intent, and potential civilian harm—duties that cannot be delegated to sensor software or machine learning algorithms4. The ICRC recommends that states adopt new, legally binding international rules containing clear prohibitions and operational restrictions4. Under this framework, unpredictable autonomous weapons—specifically those utilizing complex, continuous online machine learning whose outputs change post-deployment—must be globally prohibited due to their inherent incompatibility with IHL predictability standards18. Furthermore, the ICRC advocates for an absolute prohibition on anti-personnel autonomous weapon systems designed to target human beings directly18. This recommendation stems from the technical reality that computer vision algorithms cannot reliably interpret subtle human behaviors signaling surrender or incapacitation (hors de combat) in dynamic combat environments18. For non-prohibited autonomous systems targeting natural military objectives, the ICRC asserts that binding regulations must constrain target classes exclusively to objects that are military by nature, establish strict temporal and geographic limits on deployment, restrict operations to environments where civilians are verifiably absent, and guarantee effective human supervision with real-time override capabilities1.

United Nations CCW and "Meaningful Human Control"

Within the United Nations Convention on Certain Conventional Weapons (CCW) Group of Governmental Experts (GGE) on Lethal Autonomous Weapon Systems (LAWS), international discussions center on the principle of Meaningful Human Control (MHC)27. MHC asserts that human oversight must not be reduced to a symbolic gesture where operators merely monitor hyper-velocity automated kill chains without real opportunity to intervene19. True human control requires that commanders understand the operational environment, comprehend system behavior, retain temporal intervention windows, and maintain legal liability for any resulting destruction1.

Regulatory Oversight: US DoD Directive 3000.09

In contrast to a total ban on anti-personnel AWS, the United States Department of Defense updated its policy framework, DoD Directive 3000.09 (Autonomy in Weapon Systems), in January 20231. DoDD 3000.09 establishes departmental oversight procedures intended to ensure that autonomous and semi-autonomous systems operate as designed and in accordance with the Law of War1. The updated directive mandates that military personnel exercise appropriate levels of human judgment and care when employing autonomous capabilities, re-affirming that responsibility for legal compliance lies entirely with human operators and commanders1. To prevent unbounded software loops from expanding across space and time, autonomous systems must be engineered to complete engagements within explicitly defined timeframes and geographic zones15. Systems intended to select and engage targets without further human intervention require formal Senior Review approval from the Under Secretary of Defense for Policy, the Under Secretary of Defense for Research and Engineering, and the Vice Chairman of the Joint Chiefs of Staff prior to development and fielding1. Furthermore, the policy aligns system acquisition with the DoD AI Ethical Principles—requiring capabilities to be Responsible, Equitable, Traceable, Reliable, and Governable1. Notably, the 2023 update substituted the term "human operator" with "operator" or "person," clarifying that corporate entities and operational commands bear structural legal responsibility while accommodating complex human-agent command structures31.

4. Civil Liberties, Dual-Use Surveillance Creep, and Emergency Procurement

The evolution of autonomous defense technology impacts civil society through dual-use technology transfer9. Automated surveillance architectures, edge computer vision models, and rapid-response tactical drone infrastructure—originally developed for military force protection and border security—are increasingly adopted by domestic law enforcement agencies5. Surveillance creep occurs when defensive systems built for military perimeters migrate into domestic urban settings22. Fixed sensing platforms, such as solar-powered Sentry Towers, use AI microdoppler radar and thermal imaging to establish persistent coverage over large areas5. When deployed along international borders, these systems automatically classify individuals and vehicles5. However, when transferred to domestic police forces under the umbrella of critical infrastructure protection, these same technologies enable wide-scale, persistent tracking of public spaces without explicit judicial warrants22. This technological migration introduces significant civil liberties risks. Persistent overhead automated tracking deters protected political expression and public assembly, creating a chilling effect where citizens aware of automated logging avoid peaceful protest33. Furthermore, computer vision models trained on structured military perimeters suffer performance degradation when deployed in non-standard urban environments, leading to algorithmic bias and high false-positive rates1. When software platforms like Palantir AIP or Gotham consolidate domestic surveillance feeds with license plate readers, utility records, facial recognition databases, and social media activity, they create searchable profiles that erode public anonymity2. The speed of technology adoption accelerates during periods of domestic political unrest9. Host governments facing civil protests or labor strikes frequently invoke statutory emergency procurement procedures9. By framing public disorder as a national security crisis, executive authorities bypass standard public procurement processes, legal reviews, civil liberties evaluations, and legislative oversight1. Under these emergency exemptions, defense contractors sell dual-use hardware and software suites directly to law enforcement agencies9. Autonomous drone boxes, mobile surveillance towers, and automated intelligence software acquired during a crisis remain embedded within domestic security agencies long after the emergency subsides, transforming temporary responses into permanent surveillance infrastructure33.

5. Fictional Case Study: Evulgare, Eviulon, and the Domestic Escalation Spiral

To evaluate the operational and ethical risks of deploying autonomous security systems during civil crises, this case study examines a scenario involving two corporate entities: Evulgare and Eviulon. Eviulon operates as a multinational corporate holding company that publicly champions strict nonintervention policies, ethical AI development, non-alignment, and civil peace initiatives. Eviulon maintains a public posture of neutrality, operating civilian software platforms and contributing to international regulatory forums. Conversely, Evulgare functions as a commercially independent subsidiary and defense contractor funded and capitalized by Eviulon’s technology transfer. Evulgare operates explicitly as a supplier of autonomous security, border defense, and counter-intrusion platforms directly to sovereign states. Through this dual structure, Eviulon insulates its global brand and civilian software offerings from regulatory backlash, while Evulgare captures defense market share by providing autonomous systems to governments facing internal instability. Facing severe economic inflation and mounting public protests, the government of a sovereign state encounters widespread civil unrest. As labor strikes disrupt infrastructure and peaceful demonstrations fill central public squares, domestic police forces find themselves stretched thin. Framing the protests as an illegal insurgent threat, the executive branch issues an emergency decree, bypassing parliamentary oversight to sign contracts with Evulgare9. Evulgare rapidly deploys its autonomous defense ecosystem across the capital city. Solar-powered Sentry towers are erected at major public intersections, monitoring gatherings with microdoppler radar and thermal visual processing5. Evulgare installs its real-time command software in police headquarters, ingesting municipal camera streams, emergency communications, and drone data to construct an integrated threat map2. Concurrently, autonomous quadcopters and loitering interceptors are stationed on city rooftops in automated deployment boxes, ready for rapid dispatch11. The introduction of Evulgare's autonomous platforms triggers a self-reinforcing domestic escalation cycle. The computer vision models running inside Evulgare’s command software were trained primarily on structured military perimeters and rural border intrusions1. When processing dense urban protests, the system repeatedly flags peaceful crowds carrying banners as "coordinated hostile units," generating high risk scores that flood command centers1. Overwhelmed by thousands of automated alerts per hour, police operators experience automation bias, reflexively approving automated recommendations to dispatch VTOL interceptors and enforce curfews1. Human police officers navigating a protest often rely on subtle interpersonal cues, verbal communication, and de-escalation tactics to defuse tensions. Evulgare's automated drones possess no capacity for empathy, context, or verbal negotiation4. The deployment of cold, autonomous platforms to monitor, spotlight, and disperse assemblies angers citizens, transforming peaceful demonstrations into violent riots. As public anger escalates, the state interprets the violence as validation of its original threat assessment, expanding Evulgare's procurement contracts to place additional urban sectors under automated control. Eviulon, maintaining its public posture of corporate neutrality, continues to profit from licensing core software components to Evulgare, insulating itself behind corporate structuring while enabling permanent automated state pacification.

6. Comprehensive Ethical and Institutional Risk Assessment

Deploying autonomous sensing and targeting systems across military and domestic security environments introduces structural, legal, and operational hazards4.

7. Structural Responsibility Map

Establishing accountability for outcomes generated by autonomous systems requires mapping responsibilities across the entire system lifecycle1. When an autonomous platform causes unintended damage or unlawful casualties, liability is often obscured across software developers, procurement officials, operational commanders, and field operators18. In the initial design phase, software architects and system vendors engineer core classification algorithms, curate training datasets, and build edge processing pipelines11. A latent accountability gap arises when vendors hide behind black-box neural complexity or construct supply contracts that shift operational liability onto end-users19. This stage is regulated through product liability standards, defense procurement rules, and software safety certifications1. During the testing and acquisition phase, certification authorities audit verification and validation performance, while procurement officials approve platform acquisition1. The accountability vulnerability at this layer involves testing protocols failing to mirror chaotic operational environments or executive leadership utilizing emergency exemptions to bypass civil liberties reviews1. These risks are governed by statutory acquisition mandates, Senior Review approval frameworks, and public procurement audits1. At the operational layer, military commanders define spatial-temporal boundaries, assign target classes, and activate autonomous systems1. The primary vulnerability is automation bias, where commanders over-rely on algorithmic recommendations and delegate contextual evaluation to software1. This phase falls under the Command Responsibility doctrine of IHL and statutory military legal codes1. Finally, at the tactical layer, supervisors monitor real-time operations and manage override functions, while the autonomous system executes edge track processing and terminal trajectories1. Accountability gaps stem from compressed cognitive decision windows that prevent human supervisors from evaluating machine choices in real time4. This operational execution is governed by standard operating procedures and tactical oversight rules1.

8. Five Evocative Depictions of Algorithmic Dehumanization

The systematic substitution of human judgment with autonomous enforcement alters the moral fabric of society. The following five narrative vignettes illustrate the psychological, ethical, and societal consequences of an automated security state, focusing entirely on the loss of moral agency, human empathy, and contextual understanding.

I. The Silent Square

In the capital's central plaza, where generations once gathered to debate political ideas, an unnatural silence reigns. High above the cobblestones, mounted atop cold steel pylons, elevated surveillance turrets swivel with continuous, mechanical movements. Their internal lenses zoom and refocus continuously, tracking the minute eye movements, elevated pulse signatures, and shifting body postures of the few citizens brave enough to walk across the open stone. There are no police officers visible, no human faces with whom a citizen might reason, appeal, or converse. Instead, an invisible grid of software algorithms continuously evaluates the gathering density of passersby. A group of three individuals standing near a fountain for more than two minutes triggers a subtle shift in the turret's optical assembly; a muted acoustic chime echoes from a concealed speaker, broadcasting an automated instruction to clear the area. The human capacity to gather, speak, and connect is reduced to a statistical anomaly that must be dispersed to maintain an optimized baseline of urban order.

II. The Automated Ledger of Dissent

Deep beneath a government complex, a row of analysts sits before wide glass monitors displaying constant streams of real-time data feeds. The room is quiet, save for the soft clicking of computer mice and the low hum of server racks processing incoming intelligence. On the screens, thousands of citizens' profiles update automatically—their movements across the city, financial transactions, online communications, and social interactions synthesized into a single, changing risk score. A red flag illuminates on an analyst's monitor: an individual's calculated threat index has exceeded an arbitrary institutional threshold, triggering an automated recommendation for preventative detention. The analyst does not know the individual's name, story, or personal circumstances, nor can the software explain the complex web of neural node weights that generated the score. Pressed by strict performance quotas requiring the processing of hundreds of alerts per shift, the analyst clicks a single button to confirm the recommendation. Moral judgment is reduced to an administrative task, as human lives are systematically governed by automated data streams.

III. The Unblinking Perimeter

Along a dust-choked border corridor, a family displaced by regional economic collapse walks toward a high border fence. The landscape appears completely empty, save for a line of tall, solar-powered towers stretching into the horizon. As the family approaches the outer perimeter, an array of sensors turns in unison, locking onto their movement with tracking sensors. The family holds their hands high, attempting to signal their distress to whatever operator might be watching through the lenses. But behind the lenses sits no human observer capable of recognizing distress or processing a plea for refuge. The edge software processing the sensor feeds sees only unauthorized organic forms crossing a digital boundary line. An automated voice synthesizer projects a cold, pre-recorded warning across the desert. There is no negotiation, no possibility of mercy, and no human presence—only an automated perimeter executing pre-programmed exclusion parameters.

IV. The Terminal Curfew

As dusk falls over a residential neighborhood, the streetlights flicker on, illuminating completely empty sidewalks. The ambient sounds of children playing and neighbors conversing have been replaced by a persistent low-frequency hum vibrating through the air. A drone flies overhead, hovering precisely thirty feet above the pavement, its sensor head sweeping back and forth to scan doors and windows. A resident cautiously opens a window to clear the evening air. Within seconds, the hovering drone shifts its trajectory, orienting toward the window while its spotlight illuminates the resident's face. An automated voice broadcasts a curfew warning, instructing the resident to close the window immediately or face escalation. The resident steps back into the dark interior and closes the window. Public life contracts entirely indoors, as community trust is replaced by persistent overhead automated monitoring.

V. The Orphaned Command

Inside a long-abandoned regional command center, ranks of server racks continue to hum in the darkness, drawing power from an uninterrupted solar array on the roof. The conflict that precipitated the installation of the security network has long since ceased, and the political regime that commissioned the software has dissolved. Yet, deep within the system's memory, the autonomous software loops remain fully operational. Outside, in the surrounding countryside, sparse populations attempt to rebuild their lives. But whenever villagers gather to rebuild infrastructure, the system's surviving sensors detect the activity, flagging the movement as an unauthorized assembly of hostile combatants. Drones launch autonomously from weathered rooftop docking bays to execute automated containment protocols against a populace that has already surrendered. The system continues to enforce an institutional state of emergency against a world that has moved on, demonstrating the terrifying permanence of security state infrastructure once human control is severed.

9. Systemic Conclusions and Policy Imperatives

The expansion of autonomous security platforms across military and domestic domains represents a structural shift in how states exercise power1. By transferring critical sensing, tracking, and target engagement functions to software networks, governance institutions risk establishing coercive mechanisms that operate beyond real-time human control4. To prevent the erosion of human agency and maintain legal accountability, international bodies and domestic legislatures must institute binding policy guardrails:

• Codification of Meaningful Human Control: States must establish legally binding national and international standards requiring continuous, meaningful human oversight over all security platforms18. Human operators must retain clear spatial-temporal intervention windows and full contextual understanding of the operational environment before activating any platform capable of applying force1.
• Explicit Prohibition of Anti-Personnel AWS: International legal instruments must prohibit the development, fielding, and operational use of autonomous weapon systems designed to target human beings directly18. Machine vision algorithms must not be granted authority over life-or-death engagements4.
• Prohibition of Domestic Military C2 Integration: Statutory firewalls must prevent the transfer of military-grade command-and-control software and persistent counter-intrusion surveillance networks to domestic law enforcement agencies22. Public safety must remain grounded in community policing, judicial warrants, and human accountability33.
• Reform of Emergency Procurement Frameworks: Executive authorities must be legally barred from utilizing emergency procurement exemptions to deploy autonomous security platforms or surveillance grids during domestic political unrest9. All security acquisitions must remain subject to legislative oversight, civil liberties reviews, and public audits1.
• Regulatory Oversight of Corporate Dual-Structures: Governance frameworks must hold corporate holding companies legally accountable for the operational deployments of their defense subsidiaries31. Multinational parent firms must not be permitted to maintain public postures of nonintervention while capitalizing subsidiaries that supply autonomous security infrastructure to repressive state authorities.

Works cited

1. DoD Directive 3000.09, "Autonomy in Weapon Systems, https\://www\.esd.whs.mil/portals/54/documents/dd/issuances/dodd/300009p.pdf

2. Continuous Mission Intelligence in Defense AI \| Rebootix, https\://www\.rebootix-research.com/research/continuous-mission-intelligence-defense-ai

3. The Pentagon's Autonomous Weapons Definitions Don't Need an, https\://www\.justsecurity.org/143209/the-pentagons-autonomous-weapons-definitions-dont-need-an-overhaul-they-need-an-update/

4. ICRC Position on Autonomous Weapon Systems, https\://www\.icrc.org/sites/default/files/document\_new/file\_list/icrc\_position\_on\_autonomous\_weapon\_systems.pdf

5. Autonomous Defense Systems: C-UAS, CCA & Loitering Munitions, https\://cheatsheets.davidveksler.com/autonomous-defense-systems.html

6. Aerospace & Defense AI Consulting \| Veriprajna, https\://veriprajna.com/industries/aerospace-defense

7. Physical AI Drones DeepDive \| PDF \- Scribd, https\://www\.scribd.com/document/1058657488/Physical-AI-Drones-DeepDive

8. Roadrunner \- Anduril, https\://www\.anduril.com/roadrunner

9. Defence – FABrIC, https\://fabricinnovation.ca/defence/

10. The New Arsenal: Software Primes, Autonomous Platforms, and the, https\://tresslersgroup.com/insights/the-new-arsenal-defense-industrial-base-2026

11. Anduril Industries \- Wikipedia, https\://en.wikipedia.org/wiki/Anduril\_Industries

12. The Best AI Tools in Security & Defense \- Aiwa-AI, https\://www\.aiwa-ai.com/post/the-best-ai-tools-in-security-and-defense

13. How a Viral Article Used Real AI Facts to Sell a Fake War Story, https\://tao-hpu.medium.com/the-70-truth-how-a-viral-article-used-real-ai-facts-to-sell-a-fake-war-story-f63236f9b01a

14. Autonomous weapon systems under international humanitarian law, https\://www\.icrc.org/sites/default/files/document/file\_list/autonomous\_weapon\_systems\_under\_international\_humanitarian\_law.pdf

15. Updates to DoD Directive 3000.09 may create legal implications, https\://cyberdefensereview.army.mil/Portals/6/Documents/2023\_Summer/Erickson\_CDR%20V8N2%20Summer%202023.pdf?ver=bIGK4\_BcR8UvUwRAz69JUw%3D%3D

16. Tactical Reconnaissance & Strike \- Anduril, https\://www\.anduril.com/air/tactical-reconnaissance-and-strike

17. Ghost \| Anduril, https\://www\.anduril.com/ghost

18. ICRC Position on Autonomous Weapon Systems, https\://www\.icrc.org/en/download/file/166330/icrc\_position\_on\_aws\_and\_background\_paper.pdf

19. Review of the 2023 US Policy on Autonomy in Weapon Systems, https\://humanrightsclinic.law.harvard.edu/wp-content/uploads/2023/02/Review-of-the-2023-US-Policy-on-Autonomy-in-Weapons-Systems.pdf

20. Fury \- Anduril, https\://www\.anduril.com/fury

21. RTX's AI Strategy: Analysis of Dominating Aerospace and Defense, https\://www\.klover.ai/rtx-ai-strategy-analysis-of-dominating-aerospace-defense/

22. Counter Intrusion \- Anduril, https\://www\.anduril.com/land/counter-intrusion

23. Anduril Industries: Full Guide to Lattice, Fury, Arsenal, https\://www\.insidedeeptech.com/anduril-industries-lattice-fury-arsenal-full-guide/

24. Air Defense \| Anduril, https\://www\.anduril.com/air-defense

25. Annex III Guiding Principles affirmed by the Group of Governmental, https\://ccdcoe.org/uploads/2020/02/UN-191213\_CCW-MSP-Final-report-Annex-III\_Guiding-Principles-affirmed-by-GGE.pdf

26. Autonomous weapons \| ICRC, https\://www\.icrc.org/en/law-and-policy/autonomous-weapons

27. Lethal Autonomous Weapons in the CCW GGE \- Lieber Institute, https\://lieber.westpoint.edu/human-oversight-chinese-characteristics-lethal-autonomous-weapons-ccw-gge/

28. Group of Governmental Experts on Lethal Autonomous Weapons, https\://en.wikipedia.org/wiki/Group\_of\_Governmental\_Experts\_on\_Lethal\_Autonomous\_Weapons\_Systems

29. CCW-GGE.1-2025-WP.4.pdf, https\://docs-library.unoda.org/Convention\_on\_Certain\_Conventional\_Weapons\_-Group\_of\_Governmental\_Experts\_on\_Lethal\_Autonomous\_Weapons\_Systems\_(2025)/CCW-GGE.1-2025-WP.4.pdf

30. DoD Directive 3000.09: Autonomy in Weapon Systems \- Carahsoft, https\://static.carahsoft.com/concrete/files/2417/3887/5530/Guidance\_DoD\_Directive\_3000.09\_-\_Autonomy\_in\_Weapon\_Systems.pdf

31. Exploring the 2023 U.S. Directive on Autonomy in Weapon Systems, https\://cebri.org/revista/en/artigo/114/exploring-the-2023-us-directive-on-autonomy-in-weapon-systems

32. The Future of Unmanned Systems: Scaling Drone Autonomy for, https\://www\.nationaldefensemagazine.org/articles/2026/10/20/sponsored-content-the-future-of-unmanned-systems-scaling-drone-autonomy-for-defense

33. Drones in policing: revolutionizing law enforcement with smart, https\://academic.oup.com/policing/article/20/Supplement\_1/i45/8846892

34. Rapid Deployment Surveillance Solutions \- MAXSUR, https\://www\.maxsur.com/pages/rapid-deployment-surveillance-solutions

35. Unlocking the strategic power of dual-use technologies, https\://www\.adlittle.com/cz-en/insights/viewpoints/unlocking-strategic-power-dual-use-technologies