Glossary AI in Security and Defence

All technical terms of the article The History of AI in Security and Defence in alphabetical order. The number at the end names the section in which the term is introduced.

A

• Aegis – combat system of the US Navy, commissioned in 1983 with the cruiser Ticonderoga; combines radar, computers and missiles and in its automatic mode tracks and fires on hundreds of targets by itself, while the human sets the mode and can intervene. 2
• Agent – language model that operates tools and plans its own steps; in 2025 an agent carried out the greater part of an espionage campaign according to Anthropic, in 2026 agents found thousands of security flaws before attackers did. 8
• ARPA – Advanced Research Projects Agency, the Pentagon’s research agency, founded in February 1958 after Sputnik, DARPA since 1972; it paid for the first AI laboratories, Shakey, the Strategic Computing programme, the Grand Challenge and the dogfight in the simulator. 2
• Artificial intelligence in security and defence – use of computational methods by armed forces, intelligence services and police forces for tasks that until now needed humans: finding a target in a radar picture, assembling a situation, guiding a drone, discovering a security flaw. Introduction
• Automated data analysis – analysis of linked police databases by software that searches for patterns and relationships; according to the judgment of the Federal Constitutional Court of 16 February 2023 an interference with fundamental rights of its own, which needs a concrete danger and a law that limits data and methods. 8
• Autonomous terminal guidance – onboard computer with image recognition that follows the target marked by the pilot by itself on the last stretch when a jammer cuts the radio link; tried out in Ukraine from 2023, in series production from 2025. 7
• Autonomous weapon system – according to Pentagon Directive 3000.09 (2012) a weapon system that, once activated, selects and engages targets without further intervention by a human. 5

C

• Cybernetics – Norbert Wiener’s name (1948) for the study of control and feedback in machines and living beings, after the Greek word for helmsman; grown out of his anti-aircraft predictor of 1940. 1

D

• DART – Dynamic Analysis and Replanning Tool, planning tool for the logistics of the Gulf War of 1991, built by BBN and partners in ten weeks; drew up transport plans in minutes instead of days and counted for DARPA as proof that thirty years of AI funding had paid off. 3
• Drone – unmanned aircraft controlled by radio or satellite; the Predator flew over Bosnia in 1995 and fired the first missile in 2001. 4

E

• Expert system – programme that draws conclusions from hundreds of if-then rules of a specialist; the tool with which the Strategic Computing Initiative wanted to equip vehicle, cockpit and fleet staff. 3

F

• FPV drone – drone with first-person view: racing quadcopter for a few hundred dollars that a pilot steers with video goggles through the camera on its nose; in the war in Ukraine, with a grenade, the weapon in millions of units. 7

I

• In the loop – degree of autonomy in which the human approves every action of the machine; example the officer who chooses the target at the SAGE screen. 2

J

• JAIC – Joint Artificial Intelligence Center, office of the Pentagon founded in June 2018 that was to bundle the AI projects of all services; merged in 2022 into the Chief Digital and Artificial Intelligence Office, which has since then signed the contracts with the model makers. 6

L

• LAWS – lethal autonomous weapons systems; the term of the negotiations in Geneva since 2014 and of the resolutions of the UN General Assembly since 2023. 5
• Loitering munition – missile that waits over an area, searches for its target and strikes by itself; Harpy (1989) against radars, later Harop, Lancet and Kargu-2 against vehicles and people. 4

O

• On the loop – degree of autonomy in which the machine acts and the human watches and can abort; example Aegis in automatic mode. 2
• Out of the loop – degree of autonomy in which the human only learns after the action what happened; example Harpy, which finds and attacks a radar by itself, or the drone in the terminal approach without radio. 2

P

• Predictive policing – prediction of crimes from police data, for instance of burglaries by place and time; PredPol (2011) in Santa Cruz was the first system, and its weakness was that of the data. 4
• Project Maven – Algorithmic Warfare Cross-Functional Team of the Pentagon (April 2017), which brought learned image recognition into the analysis of drone video; continued by Palantir after Google’s withdrawal in 2018, since 2025 with language models as the Maven Smart System. 5

S

• SAGE – Semi-Automatic Ground Environment, the air defence system of the United States from 1958, with 24 direction centres, vacuum-tube computers from IBM and screens with light guns; the machine follows the tracks and guides the fighters, the human chooses the target. 1
• Shakey – robot of the Stanford Research Institute (1966 to 1972) that translated a room into symbols and planned a path in them; built with ARPA money and intended as a scout, but too slow for any deployment. 2
• Strategic Computing Initiative – DARPA programme (1983 to 1993) of about a billion dollars for an autonomous land vehicle, a co-pilot in the cockpit and a system for the fleet commander; the computers got faster, the AI goals remained unreached. 3