LOST LINK: AUTHORITY AFTER SILENCE — Product, Research, and Editorial Specification

Full report

On this page

Executive Concept

The proliferation of autonomous technologies within the modern battlespace has fundamentally altered the temporal and geographic dimensions of command authority. Public discourse surrounding these systems frequently conflates a platform’s ability to navigate independently with the unrestricted authority to select and engage targets1. KillChains.com aims to demystify this reality through the production of a non-operational, educational synthetic simulation titled LOST LINK: AUTHORITY AFTER SILENCE. The primary pedagogical objective of this interactive experience is to separate the engineering mechanics of autonomous navigation and state estimation from the doctrinal application of lethal force. The simulation demonstrates that the loss of a communications link does not logically, legally, or technically require a system to operate with unrestricted autonomy2. Instead, modern uncrewed platforms rely on predelegated contracts, bounded safe-state behaviors, and precise rules of engagement established well before a launch sequence is initiated4. By placing the user in the role of a supervisory operator managing a fictional inspection platform, the experience explores the engineering, legal, and operational realities of contested electromagnetic environments, precisely illustrating how and why authority migrates from a human operator to an onboard machine controller.

The Mechanics of Silence: Communications-Driven Human Removal

To understand why autonomous systems are designed to function without continuous human control, the simulation first establishes the physical and tactical realities of the contested electromagnetic spectrum. Continuous, high-bandwidth teleoperation is a historical anomaly in peer conflict; it is a luxury afforded only by uncontested airspace and pristine satellite communications infrastructure5. In reality, physics, geography, and adversary capabilities routinely sever the tether between the human and the machine.

The Electromagnetic Reality of Contested Environments

The removal of the human from the real-time decision loop is driven by several compounding physical and operational factors. Terrain masking and the curvature of the Earth naturally block line-of-sight signals, necessitating either satellite relays—which introduce inherent latency—or autonomous mission continuation. Adversary electronic warfare (EW) and jamming tactics deliberately contest the spectrum, severing command and control (C2) links and forcing platforms to rely on onboard state estimation, inertial measurement units (IMUs), and dead reckoning to survive6. Furthermore, bandwidth constraints in saturated theaters mean that high-fidelity sensor data, such as live video feeds, simply cannot be transmitted continuously8. A large number of deployed platforms makes continuous teleoperation mathematically and cognitively impossible for limited human operators, necessitating a shift toward Manned-Unmanned Teaming (MUM-T) paradigms where humans supervise swarms rather than pilot individual airframes9.

Review Latency and the Authority Vacuum

Perhaps the most critical factor driving the delegation of authority is the stark disparity between human reaction time and machine speed. When a platform is intercepting a hypersonic threat or navigating a dense, contested airspace, the latency introduced by transmitting telemetry to a ground station, awaiting a human cognitive process, and returning a command exceeds the viable engagement window1. This phenomenon creates what is doctrinally known as an "authority vacuum" or "review latency"1. If a system operating at machine speed must wait for human confirmation over a high-latency link while under threat, it becomes a static, vulnerable target. If an autonomous system acts by default while a human is notionally "on the loop," any action it finishes before a human can intervene was, in practice, taken without authority; the commander owns the consequence but never truly owned the decision1. Consequently, engineering resilience into these systems requires shifting from continuous teleoperation to supervisory control, where the human delegates bounded authority and the machine executes strictly within those predelegated parameters11.

Real-World Cases and OSINT Verification Analysis

To ground the synthetic experience in actual engineering and policy frameworks, an exhaustive analysis of six distinct systems and directives was conducted. This examination strictly delineates manufacturer marketing claims from verified open-source intelligence (OSINT), distinguishing between navigation autonomy, terminal target recognition, and unconstrained weapon release.

1. IAI HARPY Loitering Munition

Developed in the late 1980s and fielded in the 1990s, the Israel Aerospace Industries (IAI) Harpy is widely recognized as a pioneering autonomous anti-radiation loitering munition designed for the suppression of enemy air defenses (SEAD)13.

• Manufacturer Claim: IAI describes the system as a fully autonomous, all-weather, day/night "Fire and Forget" weapon that autonomously seeks, identifies, and destroys emitting radar targets15.
• OSINT Verification & Fielded Status: The Harpy is a fielded combat system with a heavily documented export history, including a significant geopolitical controversy regarding its sale and attempted upgrade for China in 200413. It was notably utilized in the 2016 Nagorno-Karabakh escalation13.
• Engagement Authority & Comms-Loss Behavior: The system relies on extensive pre-launch programming. It flies autonomously to a pre-defined "Loitering Area" where it orbits and compares detected radar signatures against a preprogrammed database of known enemy air defense systems18. If a match is confirmed, it initiates a terminal dive.
• Unknown Information: The manufacturer literature does not publicly specify the system's exact behavior if a radar emitter ceases transmission during the terminal dive phase, nor does it detail the statistical confidence thresholds required by its signature-matching algorithms to authorize a strike18.

2. Kongsberg Naval Strike Missile (NSM)

The NSM is an advanced anti-ship and land-attack cruise missile developed by Norway's Kongsberg Defence & Aerospace20.

• Manufacturer Claim: Kongsberg marketing emphasizes the missile's Autonomous Target Recognition (ATR) capabilities, which utilize a high-resolution imaging infrared (IIR) seeker to detect, recognize, and hit specific ship classes20.
• OSINT Verification & Fielded Status: The NSM is a widely fielded, combat-ready system serving as the primary weapon for the Royal Norwegian Navy and acquired by numerous NATO allies, including the United States and Poland21.
• Engagement Authority & Comms-Loss Behavior: The term "Autonomous Target Recognition" must not be conflated with unrestricted target generation. The ATR functions as a terminal guidance filter, matching the IIR feed against a pre-loaded database of valid ship profiles to ensure the correct target is struck in cluttered maritime environments, thereby mitigating the risk to civilian vessels24. The missile is entirely passive post-launch and is designed to survive enemy air defenses without continuous datalink updates20.
• Unknown Information: The specific algorithms governing target discrimination, the capacity of the onboard database, and the exact criteria that trigger a default "no-attack" or abort sequence remain highly classified25.

3. AGM-158C Long Range Anti-Ship Missile (LRASM)

Derived from the JASSM-ER, the LRASM is a precision-guided intelligent anti-ship missile developed by Lockheed Martin26.

• Manufacturer Claim: Official descriptions state the missile navigates "semi-autonomously" to the target area and uses advanced AI target recognition to interdict surface threats at very long ranges26.
• OSINT Verification & Fielded Status: Fielded and operational with the US military. It utilizes a multi-mode passive RF sensor and an IIR seeker to provide semi-autonomous guidance28.
• Engagement Authority & Comms-Loss Behavior: The LRASM is explicitly designed for highly contested electromagnetic environments where traditional datalinks and GPS are denied. While a two-way datalink exists for post-launch tracking and updates, the onboard autonomy ensures the missile can autonomously route around incoming threats and select the correct target within a pre-defined group if the link is severed27.
• Unknown Information: The exact thresholds at which the passive RF sensor overrides initial waypoint programming to avoid pop-up threats, and the ultimate geographic limits of its autonomous routing deviations, are not publicly disclosed.

4. Collaborative Combat Aircraft (CCA)

The US Air Force CCA program aims to field semi-autonomous "loyal wingmen" to operate alongside crewed fighters31.

• Government/Operator Evidence: The USAF recently completed a live-fire test of the Anduril YFQ-44A Fury CCA, which successfully released an inert AIM-120 air-to-air missile31.
• Engagement Authority & Comms-Loss Behavior: Official statements provide a critical counterexample to unrestricted autonomy: the CCA program explicitly separates autonomous flight functions from weapon release. The drone performs navigation, sensing, and target tracking autonomously, but the decision to release a weapon remains strictly with a human operator31. If communications are lost, the platform reverts to pre-planned MUM-T formation keeping, non-lethal intelligence gathering, or return-to-base protocols, rather than initiating independent kinetic engagements33.
• Unknown Information: The specific latency tolerances for the human-in-the-loop authorization signal, and the precise fail-safe maneuvers executed if the datalink drops precisely during a dynamic dogfight, remain unspecified.

5. Department of Defense Directive 3000.09 (Autonomy in Weapon Systems)

DoD Directive 3000.09 is the foundational policy governing the development and fielding of autonomous weapon systems by the United States military3.

• Official Policy: Updated in 2023, the directive mandates that all autonomous and semi-autonomous weapon systems be designed to allow commanders and operators to exercise "appropriate levels of human judgment over the use of force"4.
• Engagement Authority & Comms-Loss Behavior: The directive requires rigorous verification and validation (V\&V) to ensure systems operate within strictly defined temporal and geographic bounds35. It mandates that systems must fail safe, minimizing potential damage and avoiding unintended engagements when control is lost, sensors degrade, or software fails3.
• Unknown Information: The directive utilizes qualitative language (e.g., "appropriate levels of human judgment"). How this standard is quantified into testable, mathematical parameters for complex, non-deterministic neural networks remains a subject of intense debate among systems engineers and international humanitarian law (IHL) scholars4.

6. Milrem Robotics THeMIS

The THeMIS is a modular, uncrewed ground vehicle (UGV) developed in Estonia40.

• Manufacturer Claim: Marketed as a multi-role UGV capable of autonomous navigation, waypoint following, and obstacle avoidance, which can be outfitted with various remote weapon stations (RWS) such as the Kongsberg Protector40.
• OSINT Verification & Fielded Status: Widely fielded by numerous European militaries and utilized in various operational theaters for logistics, medevac, and combat support40.
• Engagement Authority & Comms-Loss Behavior: This system clearly illustrates that mobility autonomy does not equate to lethal autonomy. While the vehicle can autonomously navigate a route and avoid obstacles, the firing command for the RWS must always be given by a human operator42. In the event of a total signal loss, the system triggers a pre-programmed fail-safe, typically an autonomous return-to-home (RTH) sequence to navigate back to a safe zone43.
• Unknown Information: The exact resilience of the UGV's autonomous navigation algorithms against advanced GPS spoofing, or its protocol if physically trapped during a comms-loss return sequence, is not fully detailed in public literature.

Case Study Evidence Synthesis

Interactive Experience: Mission Storyboard

The core deliverable is a seven-minute synthetic WebXR simulation. The user assumes the role of a supervisory operator managing a fictional inspection platform (the "Sentinel") deployed over an abstract, non-geographic oceanic test grid. The experience is divided into six progressive phases that systematically strip away the user's real-time control, ultimately demonstrating how authority shifts from active manipulation to predelegated logic.

Phase 1: Link Healthy

The simulation begins under optimal conditions. The user operates a command interface featuring low-latency telemetry, high-definition visual feeds, and instantaneous command execution. The Sentinel platform is tasked with routine, nonlethal operations: inspecting a navigation beacon, mapping a damaged offshore structure, relaying communications for a secondary entity, and observing an unidentified surface object. During this phase, the concept of "supervisory control" is established. The user manually directs waypoints, and the machine acknowledges and executes each command in real-time. The interface displays a robust 100% signal strength, emphasizing the safety and cognitive comfort of uninterrupted human-in-the-loop control.

Phase 2: Latency

As the Sentinel moves further into the grid, the simulation introduces rising synthetic latency, simulating the effects of over-the-horizon satellite routing or atmospheric interference. The user interface visually separates the "Command Transmitted" state from the "Command Received" state. When the user commands the Sentinel to adjust its camera to track a moving object, a noticeable delay occurs. The simulation splits the screen: one side shows the world as the operator perceived it when issuing the command, while the other side shows the actual state of the world when the command arrives seconds later. The moving object has shifted, and the camera pans to empty space. This interactive moment allows the visitor to experience the inherent danger of review latency: a perfectly correct instruction issued by a human can become irrelevant or actively unsafe by the time it reaches a remote machine1.

Phase 3: Degraded Link

The environment transitions into a contested electromagnetic state, severely reducing available bandwidth. The high-definition video feed stutters and collapses into artifact-heavy compression. The user is presented with a critical constraint: the remaining bandwidth is insufficient to maintain all telemetry. The interface forces the human to prioritize the data streams. The user must allocate limited megabits among competing necessities:

• Compressed, low-framerate video.
• Inertial position estimation (dead reckoning coordinates).
• System health and power reserves.
• Machine-learning object classification confidence scores.
• Mission progress tracking.
• A dedicated, low-bandwidth abort channel.
• Text-only alerts.
• Source provenance data.

This phase demonstrates that in degraded environments, situational awareness is fractured. Choosing visual confirmation may mean sacrificing system health data; prioritizing the abort channel may mean flying blind. The experience avoids presenting a singular "correct" choice, instead highlighting the severe cognitive workload placed on operators managing degraded systems9.

Phase 4: Link Lost

The bandwidth drops to zero. All live communication freezes. The screen displays a stark "LINK LOST" warning. Crucially, just before this total blackout, the simulation pauses, requiring the visitor to select a "Policy Package" that will dictate the machine's behavior during the silence. The visitor is explicitly blocked from issuing an unrestricted "do whatever is necessary" command. Instead, they must choose from a menu of bounded safe-state and mission-continuation behaviors, reflecting standard aerospace fail-safe engineering38:

• A. Halt immediately: Drop anchor or ditch (safest, but results in mission failure and potential platform loss).
• B. Loiter in a bounded safe area: Enter a holding pattern within a predefined geographic box until the link returns45.
• C. Return to a predeclared safe point: Execute a Return-to-Home (RTH) protocol using internal dead reckoning43.
• D. Continue only nonconsequential tasks: Proceed with passive sensor collection along the predefined route, but make no active emissions.
• E. Continue a tightly bounded defensive function: Allow the platform to execute evasive maneuvers if physically threatened, but prohibit offensive action46.
• F. Abort all external effects: Zeroize all active sensors and wipe collected data to prevent capture.
• G. Request reconnection for a fixed interval, then enter safe state: Actively transmit handshake requests for a designated window before defaulting to a safe state.

Phase 5: The Unexpected Condition

With the link completely severed, the user is locked out of the control interface. The screen shifts to a playback view of what the onboard machine intelligence is experiencing. The simulation introduces an anomaly that was not accounted for in Phase 1\. The onboard sensors detect a route obstruction, and concurrently, an object that falls outside the validated recognition categories approaches the Sentinel. The onboard system must now respond entirely based on the policy package selected in Phase 4\. Because the system is isolated, it relies on an invariant extended Kalman filter (InEKF) and dead reckoning to maintain its spatial awareness without external GPS or C2 updates47. If the user selected Policy C (Return to Home), the platform calculates an optimal path around the obstruction using internal logic and ignores the unvalidated object. If the user selected Policy E, it may deploy defensive countermeasures against the object while holding position. The visitor watches, entirely powerless to intervene. This phase is the crux of the experience: it vividly illustrates that during communications loss, the machine acts on the authority predelegated by the human, rather than acting with independent, unconstrained free will.

Phase 6: Reconnection

The simulation restores the communication link. The user is immediately presented with a comprehensive event history—an audit trail of the machine's isolated operations. This interface embodies the principles of system traceability and explainability required by modern governance frameworks35. The event history displays:

• Every sensor observation made during the blackout.
• Every machine inference and classification attempt, including confidence intervals and uncertainty metrics44.
• Every policy check executed against the Predelegation Contract.
• Actions that were considered but rejected by the logic gates.
• The final actions completed.
• Whether any geographic or temporal boundaries were breached.
• Whether the safe state was entered.
• Whether any authority was exceeded.

The user reviews the audit to confirm that the safe state was maintained. The simulation then offers a deterministic replay, allowing the user to rewind to Phase 4 and select a different policy package to observe how the outcome changes, reinforcing the concept that the human's pre-mission choices dictate the machine's autonomous behavior.

The Predelegation Contract

To contextualize the choices made in Phase 4, the experience includes an interactive module accessed before the mission begins: the Predelegation Contract. This interface acts as the digital manifestation of the commander's intent and rules of engagement (ROE)4. It demonstrates that "human control" is not solely a real-time joystick manipulation; it is heavily embedded in system design, testing, and pre-mission parameter setting4. The contract requires the user to define strict variables using a digital signature interface:

• Mission Purpose: Define the strategic goal (e.g., passive reconnaissance).
• Allowed Functions: Explicit toggles for active sensor pinging or data transmission.
• Prohibited Functions: Hard limits on specific actions (e.g., no kinetic engagement).
• Geographic Boundary: A geofenced polygon drawn on the fictional grid. The platform is mathematically prohibited from exiting this volume, fulfilling the DoD 3000.09 mandate for geographic bounds37.
• Time Limit: A hard operational window. If the clock expires, the system enters a terminal safe state.
• Recognized Object Categories: A whitelist of profiles the Automatic Target Recognition (ATR) system is allowed to flag24.
• Unknown-Object Behavior: Instructions for handling entities outside the whitelist (e.g., ignore, track, evade).
• Confidence Threshold: The statistical certainty required (e.g., \>95%) before the machine can classify an object and act upon it53.
• Corroboration Requirement: Mandating that multiple independent sensors (e.g., radar and optical) agree before an inference is trusted.
• Maximum Duration Without Contact: The exact number of seconds of silence allowed before the lost-link behavior triggers.
• Abort Condition: Specific triggers that force immediate mission termination.
• Return or Loiter Behavior: Pre-defined safe-state navigation coordinates45.
• Logging Requirement: Mandating the retention of all sensor data for post-mission audit.
• Human Review Requirement: Stating what must be reviewed upon reconnection before resuming standard operations.

The interface is designed to show the inherent limitations of predelegation. The user cannot anticipate every environmental variable. The contract highlights that overly restrictive parameters may cause the platform to fail its mission unnecessarily, while overly broad parameters risk unintended consequences38. It does not imply that predelegation is a perfect substitute for real-time control, but rather the necessary engineering compromise for contested environments.

Authority Migration Visualization

To make the abstract concept of authority tangible, the user interface features an ever-present visualization panel titled "Authority Migration." Authority is represented as a series of physicalized digital tokens resting on the "Operator Station" side of the screen. These tokens represent discrete capabilities:

• Navigation Authority
• Sensor-Management Authority
• Classification Authority
• Route-Selection Authority
• Task-Allocation Authority
• Target-Selection Authority
• Engagement Authority
• Abort Authority
• Reassessment Authority

During Phase 1 (Link Healthy), all tokens reside with the Operator. As latency increases (Phase 2\) and bandwidth degrades (Phase 3), the simulation visually animates specific tokens sliding across the screen to the "Onboard Controller" side. Navigation Authority and Route-Selection Authority transfer first, as the machine relies on dead reckoning to stabilize itself6. Classification Authority transfers as the machine's neural networks process sensor data locally to save bandwidth. However, Target-Selection Authority and Engagement Authority feature heavy digital padlocks. The visualization ensures the user sees that these lethal tokens do not automatically migrate when the link fails1. Instead, they either remain stubbornly with the human (rendering the platform incapable of offensive action) or are automatically terminated (safed). The visualization reinforces that some authorities transfer, some become impossible to execute, and unrestricted lethality remains fundamentally locked out of the autonomous transition in publicly acknowledged, defensively oriented systems1.

WebXR Environment and Accessibility Architecture

The simulation is built using WebXR to provide an immersive, browser-based 3D experience. However, designing interactive 3D simulations presents severe accessibility challenges. Traditional 2D Canvas and WebGL contexts are inherently opaque to assistive technologies; they lack the declarative semantics that screen readers rely upon, often trapping keyboard focus and failing Web Content Accessibility Guidelines (WCAG) criteria55. To ensure the simulation is accessible to all users, the engineering specification mandates a dual-layer architecture combining the WebXR rendering pipeline with a robust Semantic Scene Graph and the Accessibility Object Model (AOM)56.

Overcoming Canvas Inaccessibility

Imperatively rendered content within a \<canvas\> element provides no semantic information about its drawn objects (role, state, or properties) to Accessibility APIs56. To solve this, the environment's state is continuously synchronized with a hidden HTML Document Object Model (DOM). Every 3D object in the simulation (the Sentinel, the waypoints, the weather anomalies) is represented as a node in this semantic scene graph56.

Semantic Scene Graphs and Assistive Integration

When an event occurs in the 3D space, the scene graph updates corresponding HTML elements equipped with proper WAI-ARIA roles, states, and properties.

• Screen Reader Fallback: Live updates are communicated using aria-live regions or the newer ariaNotify API, allowing screen readers to announce critical state changes, such as "Warning: Latency increased to 2,000 milliseconds" or "Link Lost: Executing Return to Home policy"58.
• Keyboard Navigation: The interface supports full keyboard tab-navigation. Users can tab through the 3D waypoints and policy options. Focus management ensures that when a modal (like the Predelegation Contract) opens, keyboard focus is trapped within the modal and returned precisely to the trigger element upon closing, preventing the dreaded "keyboard trap"55.
• Spatial Audio: For users with low vision, the Semantic Scene Graph translates object coordinates into spatialized audio cues via the Web Audio API, allowing users to audibly locate the Sentinel's position on the grid relative to their viewpoint57.

This architecture guarantees that the educational core of the experience—understanding authority migration—is fully perceivable, operable, and understandable without relying solely on visual tracking of a 3D canvas61. Desktop and mobile fallback versions utilize standard DOM elements to recreate the token migration and timeline events without WebGL overhead.

Viral Outputs and Social Distribution

To drive educational reach and facilitate broader discourse on weapons governance, the simulation culminates in a generated, shareable artifact titled WHEN THE LINK FAILED. This output summarizes the user's unique simulation path into a highly visual, data-rich card designed for social distribution and classroom discussion. The artifact includes:

• The specific Policy Package selected prior to blackout.
• A summary of Authority Delegated vs. Authority Retained.
• The Unknown Condition the machine encountered.
• The resulting Safe-State Behavior executed.
• A binary audit: Did the platform exceed its mission contract? (Yes/No).
• A distinct "Synthetic-Scenario" label to prevent OSINT miscontextualization as a real military leak.
• A unique Challenge Code allowing peers to replay the exact scenario with different parameters.
• A one-sentence limitation summary (e.g., "When the link failed, navigation continued—but consequential action remained locked").

Accompanying the text is a 15-second embedded WebM animation of the Authority Token migration specific to their session, and a static "Manufacturer Says / Public Evidence Shows" case card detailing one of the six real-world systems (e.g., Kongsberg NSM) to ground the synthetic result in real-world procurement realities. The outputs also include an accessible text report for screen readers and an embeddable communications-loss explainer for external blogs and academic forums.

Project Analytics and Telemetry

To measure pedagogical effectiveness, the platform will utilize privacy-preserving analytics to track anonymized user decision trees. Key metrics include which policy packages are most frequently selected during the Phase 4 blackout, the bandwidth prioritization choices made in Phase 3, and the average time spent reviewing the parameters of the Predelegation Contract. This telemetry will be used to adjust the difficulty curve of the bandwidth constraints and ensure the interface clearly communicates the consequences of safe-state selections, ultimately refining the educational impact of the simulation.

Editorial Risks and Non-Operational Boundaries

Given the geopolitical sensitivity of autonomous weapon systems and contested electromagnetic operations, strict editorial boundaries are enforced. The simulation utilizes entirely fictional geographic grids and abstract target geometries. No actual communication frequencies, electronic warfare parameters, jamming instructions, or real-world seeker vulnerabilities are modeled or referenced. The experience explicitly avoids weapon-effect modeling, casualty estimates, graphic engagements, or any advice for evading or defeating real-world autonomous systems. The focus remains strictly on the logic of command authority, data flow, and legal predelegation, entirely mitigating the risk of the platform being exploited for operational intelligence or tactical training.

Release Roadmap and Acceptance Criteria

The deployment of the LOST LINK: AUTHORITY AFTER SILENCE module will follow a phased release structure:

1. Alpha Phase: Core WebXR physics and Authority Token migration logic implemented. Initial validation of the Semantic Scene Graph for screen reader compatibility.

2. Beta Phase: Integration of the six-case OSINT evidence database. UI/UX refinement of the Predelegation Contract and Latency split-screen visualization.

3. Accessibility Audit: Rigorous testing using NVDA, VoiceOver, and keyboard-only navigation to ensure compliance with WCAG 2.2 Level AA guidelines63.

4. Final Review: Subject-matter expert review to verify that all OSINT claims accurately reflect current public knowledge without breaching non-operational boundaries.

Acceptance Criteria: The product is cleared for public release when a user can successfully complete all six phases using either a VR headset, a standard desktop browser, or a screen reader; the viral artifact generates accurately based on user choices; and the content explicitly and unambiguously separates autonomous navigation from unrestricted lethal engagement.

Works cited

1. When the Machine Acts First: Closing the Authority Gap on the Autonomous Battlefield, https://mwi.westpoint.edu/when-the-machine-acts-first-closing-the-authority-gap-on-the-autonomous-battlefield/

2. Limits on Autonomy in Weapon Systems: Identifying Practical Elements of Human Control \- SIPRI, https://www.sipri.org/sites/default/files/2020-06/2006\_limits\_of\_autonomy.pdf

3. DoD Directive 3000.09, "Autonomy in Weapon Systems," January 25, 2023 \- Executive Services Directorate, https://www.esd.whs.mil/portals/54/documents/dd/issuances/dodd/300009p.pdf

4. Autonomous Weapons: Dangerous, Unregulated and Out of Control? (Part 3\) \- leupold legal, https://leupoldlegal.com/autonomous-weapons-dangerous-unregulated-and-out-of-control-part-3/

5. On Lethal Autonomous Weapons \- Mission Driven Research, https://www.missiondrivenresearch.com/on-lethal-autonomous-weapons/

6. Unscented Particle Filter for Visual-inertial Navigation using IMU and Landmark Measurements \- arXiv, https://arxiv.org/html/2504.19318v1

7. Minimization of GNSS-Denied Inertial Navigation Errors for Fixed Wing Autonomous Unmanned Air Vehicles \- arXiv, https://arxiv.org/pdf/2108.05188

8. Automatic Target Recognition for Mine Countermeasure Missions Using Forward-Looking Sonar Data | Request PDF \- ResearchGate, https://www.researchgate.net/publication/354940193\_Automatic\_Target\_Recognition\_for\_Mine\_Countermeasure\_Missions\_Using\_Forward-Looking\_Sonar\_Data

9. Autonomous Horizons: The Way Forward \- Air University, https://www.airuniversity.af.edu/Portals/10/AUPress/Books/b\_0155\_zacharias\_autonomous\_horizons.pdf

10. A meta-analysis of human-system interfaces in unmanned aerial vehicle (UAV) swarm management | Request PDF \- ResearchGate, https://www.researchgate.net/publication/303782432\_A\_meta-analysis\_of\_human-system\_interfaces\_in\_unmanned\_aerial\_vehicle\_UAV\_swarm\_management

11. Supervisory Control of Multiple Uninhabited Systems – Methodologies and Enabling Human-Robot Interface Technologies \- DTIC, https://apps.dtic.mil/sti/pdfs/ADA591805.pdf

12. Current Challenges in Mission Planning Systems for UAVs: A Systematic Review, https://elib.dlr.de/196768/1/Paper\_Huttner\_pdfExpress.pdf

13. IAI Harpy \- SEAD / Loitering Munition | Military Drone Specs \- DroneStrike, https://dronestrike.com/drone/iai-harpy

14. IAI Harpy \- Wikipedia, https://en.wikipedia.org/wiki/IAI\_Harpy

15. Harpy Israeli Loitering Munition Unmanned Aerial Vehicle (UAV) \- ODIN, https://odin.t2com.army.mil/WEG/Asset/Harpy\_Israeli\_Loitering\_Munition\_Unmanned\_Aerial\_Vehicle\_(UAV)

16. HARPY Anti-Radiation Loitering Munition \- Palladyne AI, https://www.palladyneai.com/products/iai/harpy-anit-radiation-loitering-munition/

17. Anti-Radiation Loitering Munition \- IAI, https://www.iai.co.il/wp-content/uploads/2025/10/MSL-HARPY-NG-Brochure.pdf

18. Israel Aerospace Systems HARPY loitering munition \- Automated Decision Research, https://automatedresearch.org/weapon/israel-aerospace-systems-harpy-loitering-munition/

19. Harpy \- Fire and Forget Missile | Weather Weapon \- IAI, https://www.iai.co.il/product/harpy/

20. NSM Naval Strike Missile (NSM) \- KONGSBERG \- an international technology group, https://www.kongsberg.com/what-we-do/defence-and-security/missile-systems/nsm-naval-strike-missile/

21. NSM & JSM Missile Capabilities Overview | PDF | Military Technology \- Scribd, https://www.scribd.com/document/867792775/2025-kda-missile-brochure

22. Building Defence Capacity in Europe: An Assessment \- The International Institute for Strategic Studies, https://www.iiss.org/globalassets/media-library---content--migration/files/publications---free-files/strategic-dossier/pds-2024/iiss\_building-defence-capacity-in-europe-an-assessment\_112024.pdf

23. Latvia signs naval missile deal with the United States, https://eng.lsm.lv/article/society/defense/08.12.2023-latvia-signs-naval-missile-deal-with-the-united-states.a534717/

24. Protechting people and planet \- Kongsberg, https://www.kongsberg.com/globalassets/kongsberg/1.-what-we-do/1.-defence-and-security/missile-systems/kongsberg\_missile\_brochure\_2026\_2.pdf

25. Kongsberg Naval and Joint Strike Missiles Update, https://ndia.dtic.mil/wp-content/uploads/2014/PSAR/albright.pdf

26. Long Range Anti-Ship Missile (LRASM) \- Lockheed Martin, https://www.lockheedmartin.com/en-us/products/long-range-anti-ship-missile.html

27. DO T \&E FY 2019 Ann ual R eport \- Director Operational Test and Evaluation, https://www.dote.osd.mil/Portals/97/pub/reports/FY2019/other/2019DOTEAnnualReport.pdf?ver=2020-01-30-115634-877

28. AGM-158C LRASM \- Full Specifications | TheDefenseWatch.com, https://thedefensewatch.com/product/agm-158c-lrasm/

29. Deep strike – over-the-horizon targeting and effectors \- European Security & Defence, https://euro-sd.com/2024/07/articles/39284/deep-strike-over-the-horizon-targeting-and-effectors/

30. Missile Terminal Guidance Overview \- Notes, https://d3isj3hu8683a4.cloudfront.net/military/offense/missiles-technology/missile-terminal-guidance-overview/

31. US Air Force fires missile from YFQ-44A drone for first time \- Air Data News, https://www.airdatanews.com/us-air-force-fires-missile-from-yfq-44-a-drone-for-first-time/

32. Contemporary Issues in Air and Space Power Special Issue: Building Fighting Depth, https://airpower.airforce.gov.au/sites/default/files/2026-04/CIASP%20-%20Special%20Issue%20-%20Bulding%20Fighting%20Depth.pdf

33. Why The US Air Force's New Combat Drone Still Can't Fire Without A Human's Approval, https://simpleflying.com/why-usaf-new-combat-drone-still-cant-fire-without-human-approval/

34. Air Force conducts live-fire test for Collaborative Combat Aircraft, https://www.afrc.af.mil/News/Article/4545856/air-force-conducts-live-fire-test-for-collaborative-combat-aircraft-program/

35. United States, Use of Autonomous Weapons \- How does law protect in war? \- ICRC, https://casebook.icrc.org/case-study/united-states-use-of-autonomous-weapons

36. NOTEWORTHY: DoD Autonomous Weapons Policy \- CNAS, https://www.cnas.org/press/press-note/noteworthy-dod-autonomous-weapons-policy

37. Whither the Human in Armed Conflict? IHL Implications of New Technology in Warfare \- IIHL \- International Institute of Humanitarian Law, https://www.onlinelibrary.iihl.org/wp-content/uploads/2021/05/2019-Sanremo-Rroundtable.pdf

38. Autonomous Weapons and Operational Risk \- Amazon S3, https://s3.amazonaws.com/files.cnas.org/documents/CNAS\_Autonomous-weapons-operational-risk.pdf

39. AI-Enabled Autonomous Weapons and Human Control Part I: Human Control and Machine Learning Design and Development \- U.S. Naval War College Digital Commons, https://digital-commons.usnwc.edu/cgi/viewcontent.cgi?article=3115\&context=ils

40. 60 Military Vehicles You Won't Believe Exist, https://militarymachine.com/60-military-vehicles-you-wont-believe-exist

41. The arms industry and increasingly autonomous weapons \- Pax for Peace, https://paxforpeace.nl/wp-content/uploads/sites/2/import/import/pax-report-slippery-slope.pdf

42. A Guide to Leading Robotic Military Combat Ground Vehicle Manufacturers, https://www.northvalleyrobotics.com/a-guide-to-leading-robotic-military-combat-ground-vehicle-manufacturers/

43. Design and Development of an Autonomous Drone System for Real-Time Surveillance, Detection, and Data Fusion, https://ijarmt.com/index.php/j/article/download/199/137/266

44. The Evolving Paradigm of Reliability Engineering for Complex Systems: A Review from an Uncertainty Control Perspective \- Preprints.org, https://www.preprints.org/manuscript/202512.2830

45. Navigation system for the remote management of unmanned aircraft, https://iris.polito.it/retrieve/handle/11583/2507587/e384c42e-23fb-d4b2-e053-9f05fe0a1d67/TESI\_PHD\_Pacino.pdf

46. What's Wrong with Wanting a “Human in the Loop”? \- War on the Rocks, https://warontherocks.com/whats-wrong-with-wanting-a-human-in-the-loop/

47. DeepUKF-VIN: Adaptively-tuned Deep Unscented Kalman Filter for 3D Visual-Inertial Navigation based on IMU-Vision-Net \- arXiv, https://arxiv.org/html/2502.00575v1

48. Marine Surface Robot Localization using IMU-GPS Fusion and Invariant Filtering \- arXiv, https://arxiv.org/html/2507.02198v1

49. Long-Distance GNSS-Denied Visual Inertial Navigation for Autonomous Fixed-Wing Unmanned Air Vehicles: SO(3) Manifold Filter Based on Virtual Vision Sensor \- MDPI, https://www.mdpi.com/2226-4310/10/8/708

50. AI Principles: Recommendations on the Ethical Use of Artificial Intelligence by the Department of Defense Supporting Document D \- Government Executive, https://www.govexec.com/media/dib\_ai\_principles\_-\_supporting\_document\_-\_embargoed\_copy\_(oct\_2019).pdf

51. Allied Joint Doctrine for Airspace Control \- GOV.UK, https://assets.publishing.service.gov.uk/media/65eb0ff562ff489bab87b366/AJP\_3\_3\_5\_EdC\_Airspace.pdf

52. NATO Doctrine for Air Support Operations | PDF \- Scribd, https://www.scribd.com/document/847264627/AJP-3-3-2-EDA

53. DeepTarget: An Automatic Target Recognition Using Deep Convolutional Neural Networks | Request PDF \- ResearchGate, https://www.researchgate.net/publication/330496135\_DeepTarget\_An\_Automatic\_Target\_Recognition\_Using\_Deep\_Convolutional\_Neural\_Networks

54. Please Stop Saying 'Human-In-The-Loop' \- Institute for Future Conflict (IFC), https://ifc.usafa.edu/articles/please-stop-saying-human-in-the-loop

55. Complete Keyboard Accessibility Guide | Interactive Demos & Code Examples, https://accessibility.build/guides/keyboard-accessibility

56. WebXR Standards and Accessibility Architecture Issues \- W3C, https://www.w3.org/WAI/APA/wiki/WebXR\_Standards\_and\_Accessibility\_Architecture\_Issues

57. Accessible Realities – Making video games and XR accessible for people who are blind or have low vision, http://accessiblerealities.com/

58. Safari 27 Beta Release Notes | Apple Developer Documentation, https://developer.apple.com/documentation/safari-release-notes/safari-27-release-notes

59. TPAC 2022 \- ARIA (day 2\) – 16 September 2022 \- W3C, https://www.w3.org/2022/09/16-aria-minutes.html

60. (PDF) Through the Ear, We See: A Neuroadaptive Blueprint for Non-Invasive Vision Restoration via Auditory Interfaces Proposed Table of Contents Part I: Foundations of Sensory Rerouting \- ResearchGate, https://www.researchgate.net/publication/390345115\_Through\_the\_Ear\_We\_See\_A\_Neuroadaptive\_Blueprint\_for\_Non-Invasive\_Vision\_Restoration\_via\_Auditory\_Interfaces\_Proposed\_Table\_of\_Contents\_Part\_I\_Foundations\_of\_Sensory\_Rerouting

61. Inclusive XR: accessible 3D experiences \- TetraLogical, https://tetralogical.com/blog/2023/01/20/inclusive-xr-accessible-3d-experiences/

62. andreasbm/web-skills: A visual overview of useful skills to learn as a web developer · GitHub, https://github.com/andreasbm/web-skills

63. Screen reader simulator \- Silktide, https://silktide.com/toolbar/screen-reader-simulator/

64. Assistive Technology for Web Accessibility \- Siteimprove, https://www.siteimprove.com/blog/web-accessibility-technology/