Terminology atlas
These are bounded editorial definitions used by KillChains.com. Where governments, international organizations, manufacturers, and researchers use a term differently, the definition identifies the site’s analytical scope rather than claiming universal agreement.
Answer-first summary
Terms such as autonomous, human-on-the-loop, and meaningful human control are not used identically by every institution, so each definition states the bounded meaning used on KillChains.com.
Read the supporting sectionAutonomy of a specific function—such as navigation, tracking, or recognition—is different from authority to select and engage a target.
Read the supporting section“AI-enabled,” “automatic,” “autonomous,” “human-supervised,” and “fire-and-forget” answer different questions. A system can navigate independently while requiring human weapon release; a defensive system can automatically engage incoming materiel without using machine learning; a human can remain formally present while lacking a meaningful review window.
Core models
A family of models describing how an adversary progresses from knowledge or access to manipulating an AI system’s data, model, context, tools, infrastructure, memory, or users and then creates an impact. The term is not one universal standard.
Link to this termAutonomy and AI
A chain in which AI or machine learning assists, automates, or autonomously performs at least one consequential function. It is not necessarily an autonomous weapon.
Link to this termHuman control
The U.S. Department of Defense policy standard requiring autonomous and semi-autonomous weapon systems to allow commanders and operators to exercise context-appropriate judgment over the use of force.
Link to this termF2T2EA stages
Estimate whether the intended outcome occurred, what uncertainty remains, whether the track persists, and whether another action or investigation is justified.
Link to this termAutonomy and AI
Software compares sensed features with programmed or learned categories to recognize or discriminate an object. ATR does not by itself establish unrestricted target generation, legal validation, or authority to engage.
Link to this termHuman factors
Over-reliance on automated advice, including commission errors from following incorrect recommendations and omission errors from failing to act when automation does not alert.
Link to this termAutonomy and AI
A system determines or adjusts its movement without continuous steering commands. It does not imply authority to classify, select, or engage a target.
Link to this termAutonomy and AI
A weapon that, once activated, can select and engage targets without further operator intervention under the functional definition used by the U.S. Department of Defense. Other institutions use related but nonidentical language and regulatory positions.
Link to this termAutonomy and AI
Independent performance of a bounded task such as navigation, sensor management, tracking, recognition, route selection, or assessment. It must be distinguished from authority over the use of force.
Link to this termEvidence and models
Several apparently separate reports trace back to the same underlying observation or claim, creating the illusion of independent corroboration.
Link to this termSimulation
A comparison between the user’s chosen intervention and a predefined alternative while preserving the same underlying scenario evidence and timing.
Link to this termCore models
Lockheed Martin’s intrusion-oriented sequence: Reconnaissance, Weaponization, Delivery, Exploitation, Installation, Command and Control, and Actions on Objectives.
Link to this termSimulation
Replaying the same scripted observations, timing, and scenario seed so that different decisions or policies can be compared without hiding the consequences behind new randomness.
Link to this termF2T2EA stages
Apply an authorized effect. On KillChains.com this may be shown abstractly as warning, containment, interruption, interception, or no action rather than weapon-employment procedure.
Link to this termEvidence and models
The strength of support for a claim, considering source reliability, independence, provenance, freshness, relevance, contradiction, directness, and the conditions under which the information was produced.
Link to this termEvidence and models
A visible classification of how a claim is supported, such as officially documented, independently corroborated, manufacturer-described, alleged, disputed, or publicly unspecified.
Link to this termCore models
Find, Fix, Track, Target, Engage, Assess: a familiar military dynamic-targeting sequence used on this site as a high-level analytical model.
Link to this termF2T2EA stages
Detect or discover a potentially relevant object, signal, event, or activity through sensing, intelligence, or observation.
Link to this termF2T2EA stages
Establish sufficient confidence in location, identity, observation quality, and relevance for continued tracking and evaluation.
Link to this termHuman control
An affirmative human action is required before the system performs a designated consequential function for a particular engagement or action.
Link to this termHuman control
A system performs delegated functions automatically while a human supervises and may have authority and time to intervene, abort, override, or deactivate.
Link to this termHuman control
After activation, no target-specific human approval is required for the system to select and engage within the delegated envelope. Earlier human design, policy, mission, and activation decisions remain part of the causal architecture.
Link to this termCore models
A staged model connecting an initial observation, access, or preparation step to a later objective. Its defensive value is the identification of dependencies and interruption points.
Link to this termCore models
A distributed network in which sensing, fusion, command, and effect functions can be composed across many nodes and rerouted rather than proceeding through one fixed linear path.
Link to this termHuman control
A communications condition in which live command or telemetry is unavailable. It does not determine its own meaning: the system may halt, loiter, return, continue reversible tasks, or follow another predeclared policy.
Link to this termHuman control
A normative concept emphasizing sufficient information, understanding, predictability, authority, time, and intervention capability for human judgment. It is influential in policy debate but not one universally codified treaty standard.
Link to this termEvidence and models
A score produced by a model under its data, categories, calibration, and current inputs. It is not automatically evidence quality, positive identification, legal correctness, or ground truth.
Link to this termHuman factors
A person is formally assigned an approval or supervisory role but lacks the time, information, authority, interface, or functioning intervention path needed to exercise independent judgment.
Link to this termHuman control
A human-defined permission established before the event that allows a system to perform a bounded action when specified evidence, mission, geographic, temporal, and safety conditions are satisfied.
Link to this termEvidence and models
Information about where data, a claim, model, document, or inference came from and how it changed. Provenance helps detect circular reporting, stale inputs, tampering, and unsupported summaries.
Link to this termEvidence and models
Reviewed public evidence does not establish the relevant operating mode, authority, threshold, configuration, or status. Unknown information must not be silently converted into either human-controlled or autonomous.
Link to this termHuman control
A predefined, bounded system condition entered when assumptions, communications, sensors, authority, or operating limits fail. Safe for the platform does not necessarily mean risk-free for every affected party.
Link to this termEvidence and models
Combining observations from multiple sensors or reports into a track, estimate, or shared picture while accounting for timing, uncertainty, source dependence, and contradiction.
Link to this termSimulation
The browser submits bounded user intent, while PHP validates the command and determines the official stage, evidence visibility, outcome, sequence, score, and debrief state.
Link to this termSimulation
A fictional setting, event stream, evidence set, timing, score, and outcome created solely to teach structure, uncertainty, control, and causality. It is not evidence about a real system.
Link to this termF2T2EA stages
Validate and prioritize a candidate, evaluate applicable constraints, select a course of action or effector, and determine the required authority. It is not merely an AI classification label.
Link to this termWeb technology
The JavaScript 3D rendering library used for the site’s WebGL and WebXR visualizations, with semantic HTML and Canvas fallbacks for essential content.
Link to this termF2T2EA stages
Maintain continuity of identity and movement over time while accounting for uncertainty, missed observations, and competing association hypotheses.
Link to this termWeb technology
A web standard that gives compatible browsers access to immersive virtual- and augmented-reality devices. On KillChains.com it is progressive enhancement rather than a content requirement.
Link to this termUse and citation
When a term is contested, cite both the KillChains.com definition and the underlying official or institutional source. Do not remove qualifiers such as “after activation,” “operator-supervised,” “manufacturer-described,” or “publicly unspecified.”