AI & Cybersecurity Glossary
Comprehensive terminology reference for artificial intelligence, machine learning, and cybersecurity. Supporting N.Tellgencya training programs for Northern Territory professionals.
A
- A* Search Pronounced 'A-star'. A graph traversal and pathfinding algorithm used in many fields of computer science due to its completeness, optimality, and optimal efficiency.
- Abductive Logic Programming (ALP) A high-level knowledge-representation framework for solving problems declaratively based on abductive reasoning. Extends normal logic programming by allowing some predicates to be incompletely defined.
- Abductive Reasoning (Abduction) A form of logical inference which starts with an observation or set of observations then seeks to find the simplest and most likely explanation. Unlike deductive reasoning, it yields a plausible conclusion but does not positively verify it.
- Ablation The removal of a component of an AI system. An ablation study aims to determine the contribution of a component by removing it and analysing the resultant performance.
- Abstraction The process of removing physical, spatial, or temporal details in the study of objects or systems to more closely attend to other details of interest.
- Access Control The process of granting or denying requests for access to systems, applications and data. Can also refer to access control for facilities.
- Access Cross Domain Solution A system permitting access to multiple security domains from a single client device.
- Accountable Material Material requiring the strictest control over its access and movement, including TOP SECRET data, some caveated data, and any data designated as accountable by its originator.
- Action Selection A way of characterising the most basic problem of intelligent systems: what to do next. Associated with intelligent agents and animats—artificial systems exhibiting complex behaviour in an agent environment.
- Activation Function In artificial neural networks, the function of a node that defines the output of that node given an input or set of inputs.
- Actuators Mechanical or electromechanical devices that convert energy into motion or physical action, essential components in robotics and automated systems.
- Admissible Heuristic In pathfinding algorithms, a heuristic function that never overestimates the cost of reaching the goal. The cost it estimates is not higher than the lowest possible cost from the current point.
- Adversarial Attack Techniques that attempt to deceive machine learning models through malicious input, exploiting vulnerabilities in how models process data.
- Affective Computing The study and development of systems and devices that can recognise, interpret, process, and simulate human affects. An interdisciplinary field spanning computer science, psychology, and cognitive science.
- Agent Architecture A blueprint for software agents and intelligent control systems, depicting the arrangement of components. Architectures implemented by intelligent agents are referred to as cognitive architectures.
- Agentic AI AI systems capable of autonomous action, decision-making, and interaction with external systems without constant human oversight. While offering productivity benefits, agentic AI significantly expands the attack surface.
- Aggregation (of data) A term used to describe compilations of data that may require a higher level of protection than their component parts.
- AI Accelerator A class of microprocessor or computer system designed as hardware acceleration for artificial intelligence applications, especially neural networks, machine vision, and machine learning.
- AI-Complete (AI-Hard) Problems whose difficulty is equivalent to solving the central artificial intelligence problem—making computers as intelligent as people.
- AI Hallucination When an AI model generates false information presented as fact, including fabricated statistics, non-existent events, or plausible-sounding but incorrect technical details.
- AI/ML Bill of Materials (BOM) A comprehensive inventory of all components, dependencies, and data sources used in an AI/ML system, similar to a software BOM but specific to machine learning assets.
- AI Use Statement A transparent disclosure of how AI tools were used in creating academic or professional work, including which tools, for what purposes, and what human oversight was applied.
- Algorithm An unambiguous specification of how to solve a class of problems. Algorithms can perform calculation, data processing, and automated reasoning tasks.
- AlphaGo A computer program developed by Google DeepMind that plays the board game Go. In 2015, it became the first program to beat a human professional Go player without handicaps on a full-sized board.
- Anomalous Patterns Unusual or irregular patterns in data that deviate from expected behaviour, often indicating potential security threats, system failures, or areas requiring investigation.
- Application Control An approach in which only an explicitly defined set of trusted applications are allowed to execute on systems.
- Application Programming Interface (API) A set of subroutine definitions, communication protocols, and tools for building software. Provides clearly defined methods of communication among various components.
- Artificial General Intelligence (AGI) A type of AI that matches or surpasses human cognitive capabilities across a wide range of cognitive tasks.
- Artificial Intelligence (AI) Intelligence demonstrated by machines, in contrast to natural intelligence displayed by humans and animals. The study of 'intelligent agents': devices that perceive their environment and take actions to achieve goals.
- Artificial Neural Network (ANN) A computing system inspired by biological neural networks. It consists of interconnected nodes (artificial neurons) organised in layers that process and transmit information.
- ASCII Smuggling A prompt injection technique using special Unicode characters that appear invisible or render as whitespace to embed malicious instructions, bypassing content filtering.
- Asset Anything of value, such as ICT equipment, software or data.
- Astroturfing The practice of creating a deceptive impression of grassroots support for a policy, product, or opinion, when such support is artificially generated through coordinated campaigns.
- ATSILIRN Protocols Guidelines developed by Aboriginal and Torres Strait Islander Library, Information and Resource Network for libraries and archives managing culturally restricted materials and sacred knowledge.
- Attack Surface The amount of ICT equipment and software used in a system. The greater the attack surface the greater the chances of malicious actors finding an exploitable vulnerability.
- Attention Mechanism A machine learning technique that calculates 'soft' weights for each word (or its embedding) in the context window. It can operate in parallel (transformers) or sequentially (recursive neural networks).
- Authentication Verifying the identity of a user, process or device as a prerequisite to allowing access to resources in a system.
- Authentication Header A protocol used in Internet Protocol Security (IPsec) that provides data integrity and data origin authenticity but not confidentiality.
- Australian Government AI Technical Standard Mandatory requirements for federal agencies deploying AI systems, covering adversarial testing, security monitoring, bias testing, incident handling, and output controls.
- Australian Privacy Principles (APPs) Thirteen principles under the Privacy Act 1988 governing how organisations collect, use, disclose, and secure personal information.
- Automated Machine Learning (AutoML) A field of machine learning that aims to automatically configure an ML system to maximise its performance without requiring extensive manual tuning.
- Automata Theory The study of abstract machines and automata, as well as the computational problems that can be solved using them. A theory in theoretical computer science and discrete mathematics.
- Autonomous Robot A robot that performs behaviours or tasks with a high degree of autonomy, capable of sensing its environment and making decisions without continuous human guidance.
B
- Backpropagation A method used in artificial neural networks to calculate gradients needed for calculating network weights. Works by calculating the error at the output and distributing it backwards through the network's layers.
- Bag-of-Words Model A simplifying representation used in natural language processing where a text is represented as an unordered collection (bag) of its words, disregarding grammar and word order but keeping multiplicity.
- Batch Normalisation A technique for improving the performance and stability of artificial neural networks by normalising the inputs of each layer to have zero mean and unit variance.
- Bayesian Programming A formalism and methodology for specifying probabilistic models and solving problems when less than necessary information is available.
- Bias-Variance Tradeoff In machine learning, the property whereby models with lower bias in parameter estimation have higher variance of parameter estimates across samples, and vice versa.
- Big Data Data sets that are too large or complex for traditional data-processing application software to adequately handle. Characterised by volume, velocity, and variety.
- Big O Notation A mathematical notation that describes the limiting behaviour of a function when the argument tends towards a particular value or infinity.
- Binary Tree A tree data structure in which each node has at most two children, referred to as the left child and the right child.
- Blackboard System An artificial intelligence approach where a common knowledge base (the 'blackboard') is iteratively updated by diverse specialist knowledge sources, working together to solve complex problems.
- Boltzmann Machine A type of stochastic recurrent neural network and Markov random field, serving as the stochastic, generative counterpart of Hopfield networks.
- Bootstrap Aggregating (Bagging) A machine learning ensemble metaheuristic for primarily reducing variance by training multiple models independently and averaging their predictions.
- Bootstrapping In statistics, a resampling technique involving repeatedly sampling with replacement from a dataset. In broader context, refers to a self-starting process that proceeds without external input.
- Boosting A machine learning ensemble metaheuristic for primarily reducing bias by training models sequentially, with each model correcting the errors of its predecessor.
C
- C2PA (Coalition for Content Provenance and Authenticity) An industry standard for embedding provenance metadata in digital content, backed by Adobe, Microsoft, Google, and major camera manufacturers. Being fast-tracked as an ISO standard.
- Camera Verify Sony's implementation of C2PA for press photographers (launched June 2025), enabling cryptographic signing of images at capture to prove authenticity.
- Capsule Neural Network (CapsNet) A type of artificial neural network designed to better model hierarchical relationships, attempting to more closely mimic biological neural organisation.
- CARE Principles A framework for Indigenous data governance: Collective benefit, Authority to control, Responsibility, Ethics (Indigenous rights as primary concern).
- CCPA (California Consumer Privacy Act) California state law regulating how businesses handle consumer personal information, providing data privacy rights to California residents.
- Citation Hallucination A specific form of hallucination where AI fabricates academic references—creating fake author names, journal titles, or paper titles that appear legitimate but don't exist.
- Communications Security The controls applied to protect telecommunications from unauthorised interception and exploitation, as well as ensure the authenticity of such telecommunications.
- Confidentiality The assurance that data is disclosed only to authorised entities.
- Content Credentials Metadata attached to digital content providing information about its origin, creation process, and any modifications made. A tamper-evident seal showing the content's history.
- Context Window The maximum amount of text or information that an AI model can process and consider at one time when generating a response. Represents the AI's 'working memory,' typically measured in tokens.
- Convolutional Neural Network (CNN) A class of deep neural network most commonly applied to image analysis, using convolutional layers to automatically learn spatial hierarchies of features.
- Cross Domain Solution A system capable of implementing comprehensive data flow security policies with a high level of trust between two or more differing security domains.
- Cross-Modal Reasoning The ability to connect information across different modalities (text, images, audio) to draw conclusions that require integrating multiple types of evidence.
- Cross-Prompt Injection Attack (XPIA) An indirect prompt injection attack where malicious instructions are embedded in external content (emails, documents, web pages) that an AI system processes.
- Crossover (Recombination) In genetic algorithms, a genetic operator used to combine the genetic information of two parents to generate new offspring, analogous to sexual reproduction.
- Cryptographic Algorithm An algorithm used to perform cryptographic functions, such as encryption, integrity, authentication, digital signatures or key establishment.
- Cryptographic Hash An algorithm which takes as input a string of any length and generates a fixed length string as output. Designed to make it computationally infeasible to find inputs mapping to a given digest.
- Cryptographic Protocol An agreed standard for secure communication between two or more entities to provide confidentiality, integrity, authentication and non-repudiation of data.
- Cyber Resilience The ability to adapt to disruptions caused by cyber security incidents while maintaining continuous business operations, including the ability to detect, manage and recover.
- Cyber Security Measures used to protect the confidentiality, integrity and availability of systems and data.
- Cyber Security Event An occurrence of a system, service or network state indicating a possible breach of security policy, failure of safeguards or a previously unknown relevant situation.
- Cyber Security Incident An unwanted or unexpected cyber security event, or series of events, that has either compromised business operations or has a significant probability of doing so.
- Cyber Threat Any circumstance or event with the potential to harm systems or data.
Z
- Zero-Click Attack A cyberattack that requires no user interaction to execute. In AI systems, attacks where simply processing malicious content triggers exploitation.
- Zero-shot Learning (ZSL) A problem setup where, at test time, a learner observes samples from classes not seen during training and must predict the correct class.