What can this FAQ answer?
It gives concise, source-linked answers to the most common definitional, technical, legal, evidence, privacy, and simulation questions raised by KillChains.com.
Read the supporting sectionAnswer library
Concise answers preserve the site’s evidence labels and uncertainty. Open each answer for supporting source records and a direct path into the full explanatory page.
Answer-first summary
It gives concise, source-linked answers to the most common definitional, technical, legal, evidence, privacy, and simulation questions raised by KillChains.com.
Read the supporting sectionNo. They summarize public evidence for education and preserve uncertainty; they are not targeting guidance, legal advice, or a current operational assessment.
Read the supporting sectionThe answer comes first. Supporting links then show the relevant full explanation and public source records. Where the public record does not establish an operating mode, authority arrangement, or deployment fact, the answer says so instead of filling the gap with inference.
A kill chain is a staged model that connects an initial observation, access, or preparation step to a later operational objective. Its defensive value is that analysts can identify where the sequence depends on a link and interrupt it before the objective is reached.
F2T2EA means Find, Fix, Track, Target, Engage, and Assess. It is a familiar military dynamic-targeting formulation. KillChains.com expands normally hidden questions inside “Target,” including identification, validation, prioritization, authority, and applicable constraints.
The Lockheed Martin Cyber Kill Chain organizes an intrusion into Reconnaissance, Weaponization, Delivery, Exploitation, Installation, Command and Control, and Actions on Objectives. It is an ordered teaching model; MITRE ATT&CK provides a more granular behavior knowledge base that is not constrained to one linear order.
The phrase is not one universally standardized model. On this site, an AI-system attack chain describes how an adversary can Recon, Poison, Hijack, Persist, and create Impact in an AI-enabled application, with an Iterate or Pivot loop for agentic systems. MITRE ATLAS and NIST adversarial-machine-learning terminology supply complementary detail.
No. AI can assist imagery analysis, sensor fusion, classification, prioritization, route planning, or assessment while a human retains authority over weapon release. The correct unit of analysis is the function and context: what is automated, under which mission, target, geographic, temporal, and intervention constraints?
On KillChains.com, human-in-the-loop means an affirmative human decision is required before a particular consequential action. Human-on-the-loop means a system can act within delegated bounds while a human supervises and may intervene. Human-out-of-the-loop after activation means the system can select and act without further target-specific approval. These labels are site definitions; institutional terminology varies.
A person needs enough time, relevant and comprehensible information, a realistic ability to question or reject the recommendation, authority to delay or abort, awareness of system limits, and records that make the decision reconstructable. A nominal approval button does not establish those conditions by itself.
The main pressures are reaction time, simultaneous track volume, repetitive calculation, communications delay or loss, and the inability of one person to directly control many platforms or data streams. Removing people from a micro-decision can move their role earlier into policy, configuration, activation, supervision, exception handling, and review.
Upstream systems can decide which observations become tracks, which sources are fused, what is hidden below a threshold, how candidates are ranked, and which alternatives appear on the operator’s screen. The downstream human may retain formal authority while operating inside a machine-curated choice set.
Automation bias is over-reliance on automated advice. It can produce commission errors, where a person follows an incorrect recommendation, and omission errors, where a person fails to act because the system did not alert. High workload, repeated apparent reliability, opaque evidence, and coercive defaults can make the bias worse.
Public sources establish a highly automated defensive chain involving detection, tracking, trajectory and impact prediction, selective defense, and multi-target handling. They do not establish one universal human-authorization arrangement for every operator, period, configuration, and operating mode. KillChains.com therefore labels the control arrangement as configurable or publicly unspecified where appropriate.
The U.S. Navy publicly describes Phalanx as a self-contained system that automatically detects, evaluates, tracks, engages, and performs kill assessment against bounded incoming threats. That establishes automated local-defense capability, not that every deployment or engagement uses the same mode, nor that its core logic necessarily uses machine learning.
Project Maven and the Maven Smart System apply computer vision, data integration, and decision support to help analysts process large amounts of information. Public descriptions support AI-assisted sensing and targeting workflows; they do not establish that Maven independently authorizes weapon release.
Autonomous target selection and engagement are publicly documented at high confidence in narrowly bounded defensive or anti-materiel roles, such as local point defense and anti-radiation target-profile matching. The reviewed public record does not establish routine, open-ended autonomous selection and killing of people under standing doctrine at high confidence. Individual modes and incidents are frequently undisclosed or contested.
No. A platform can continue navigation, maintain a track, return, loiter, or guide toward a target selected before the link failed. Proof of autonomous target selection requires evidence that the system chose the particular target after activation from among candidates, not merely that it continued without a datalink.
A chain emphasizes an ordered dependency path. A kill web distributes sensing, fusion, command, and effects across many connected nodes and can reroute when one path fails. The web may be more resilient to node loss while also propagating a plausible but incorrect state more quickly.
No. A score is conditional on a model, available inputs, calibration, allowed categories, and operating conditions. Evidence quality depends on source reliability, independence, freshness, provenance, contradiction, and relevance. Neither variable alone establishes identity, legality, intent, or ground truth.
An evidence state records how a claim is supported: for example, officially documented, independently corroborated, manufacturer-described, credibly reported, alleged, disputed, operational status unspecified, control mode unspecified, or outdated. It prevents the interface from presenting unequal evidence as equally certain.
The names, places, timings, confidence scores, evidence objects, outcomes, and decision branches were created for education. They are not calibrated estimates of a real system, battle, organization, or person and cannot be used as operational evidence.
No. Three.js and WebXR progressively enhance the experience, but server-rendered HTML and local Canvas fallbacks preserve the essential content and controls for desktop, mobile, keyboard, touch, reduced-motion, and no-WebGL use.
The current public release has no accounts, ads, analytics service, server-side persistent learning profile, uploads, or collection of raw headset pose, gaze, hand, voice, room, or location data. Some challenge state uses a temporary PHP session, and optional progress can remain locally in the browser.
No. The site prohibits real targets, exact operational coordinates, casualty modeling, weapon-performance optimization, executable malware, exploit commands, credential material, arbitrary uploads, and outside-system actions. It is an educational research and simulation resource only.
Identify the exact page and claim, supply a stronger source or explain the methodological issue, distinguish factual correction from interpretive disagreement, and note whether the issue affects other records. The configured editorial contact is shown on the methodology and editorial-policy pages.
Terms
The glossary separates function-level autonomy, engagement authority, evidence quality, confidence, provenance, and simulation terms.
Open the glossaryEvidence
The source library exposes publisher, source class, date, bounded support statement, and stable source ID.
Open the source libraryCorrections
Use the editorial policy and methodology to distinguish a factual correction from an interpretive disagreement or changed source state.
Review the correction path