136terms · Plain English · Australian context

The AI Glossary.Every term, in plain English.

From foundations to governance, practical tools to the rules that apply in Australia. Written for board members weighing up AI risk, SME owners adopting their first AI tool and anyone building smarter habits. No unnecessary jargon.

136 terms across 8 sections

Section 01

AI Foundations

The building blocks. Whether you are briefing a board, onboarding a team or just getting started, this is where to begin.

Algorithm: A set of rules or instructions that a computer follows to solve a problem or complete a task. In AI, algorithms are the mathematical processes that allow systems to learn patterns, make predictions and generate outputs from data.
Artificial Intelligence (AI): The broad field of computer science concerned with building systems capable of performing tasks that typically require human intelligence. That includes learning from data, recognising patterns, understanding language, making decisions and generating content.
Artificial General Intelligence (AGI): A hypothetical AI system with a wide range of cognitive abilities comparable to humans. It would learn, reason, adapt to new situations and apply intelligence across any domain. No such system exists today. All current AI is narrow (task-specific) rather than general.
Attention Mechanism: An AI technique that lets models focus on the most relevant parts of input data when processing information. Attention is the main innovation behind transformer models and modern large language models.
Benchmarking (AI): The process of evaluating and comparing AI models using standardised tests to gauge performance and capability. Benchmarks guide organisations in choosing between models and tracking improvements over time.
Big Data: The huge volumes of structured and unstructured data generated daily from sources including social media, sensors, transactions and business systems. Big data provides the raw material many AI systems learn from.
Computer Vision: A branch of AI that lets machines interpret and understand visual information from images and video. Applications include object detection, facial recognition, optical character recognition (OCR) and quality inspection.
Deep Learning: A subset of machine learning that uses neural networks with multiple layers to learn complex patterns from large volumes of data. Powers image recognition, speech understanding, natural language processing and generative AI.
Embeddings: Numerical vector representations of data (text, images or other content) that capture meaning in a format AI models can process. Similar items produce similar embeddings, which is what makes semantic search and recommendations work.
Few-Shot Learning: A machine learning approach where models learn to perform new tasks from just a few labelled examples, typically five or fewer. Useful when labelled data is scarce.
Machine Learning (ML): A subset of AI in which systems learn patterns and make predictions or decisions from data, without being explicitly programmed for each scenario. Covers supervised learning (from labelled examples), unsupervised learning (finding hidden patterns) and reinforcement learning (learning through feedback).
Narrow AI (Weak AI): AI systems that are highly capable within a specific, limited domain but cannot generalise to other tasks. All current AI systems are narrow AI. Contrasts with the hypothetical Artificial General Intelligence (AGI).
Natural Language Processing (NLP): A branch of AI focused on getting computers to understand, interpret and generate human language. Sits behind chatbots, sentiment analysis, document summarisation, translation and search.
Natural Language Generation (NLG): A subfield of NLP focused on AI systems that produce natural, human-readable language as output. Writing a report, summarising a meeting or drafting an email are all NLG tasks.
Neural Network: A computing architecture inspired by the human brain, consisting of layers of interconnected nodes (neurons) that process information. Neural networks learn by adjusting the strength of connections between nodes based on training data. The foundation of deep learning and modern AI.
Overfitting: A problem where an AI model performs very well on its training data but poorly on new, unseen data, because it has memorised examples rather than learning general patterns. A main reason why models need to be tested on separate data before deployment.
Reinforcement Learning: A type of machine learning where an AI agent learns to make decisions by taking actions and receiving rewards or penalties based on outcomes. Powers applications such as game-playing AI, robotics and recommendation systems.
Supervised Learning: A type of machine learning in which a model is trained on labelled examples to make predictions on new, unseen data. For example, training a model on thousands of labelled customer service emails to automatically classify new ones.
Training Data: The dataset used to teach an AI model to recognise patterns and make predictions. The quality, diversity and representativeness of training data directly shapes what the model learns and where its biases may lie.
Transfer Learning: A technique where a model trained on one task is reused as the starting point for a different but related task. Transfer learning dramatically reduces the data and compute required to build effective models.
Transformer: A neural network architecture that uses a mechanism called self-attention to process data in parallel rather than sequentially. Transformers are the foundation of modern large language models and have reshaped NLP, computer vision and generative AI since 2017.
Unsupervised Learning: A type of machine learning in which a model is trained on unlabelled data to find patterns or groupings. Used for tasks like customer segmentation, anomaly detection and topic modelling.
Zero-Shot Learning: A technique in which a machine learning model can recognise or classify new concepts without having seen any labelled examples for that category. Contrasts with few-shot learning, which requires a small number of examples.

Section 02

Generative AI and Language Models

The technology behind the AI tools most people use today. ChatGPT, Claude, Copilot, Gemini and the rest.

Chatbot: A user-facing interface that allows users to ask questions and receive answers. Ranges from simple, pre-written response systems to fully conversational AI that resolves complex queries. Chatbots powered by LLMs are far more capable than earlier rule-based versions.
Context Window: The maximum amount of text (measured in tokens) that a large language model can process in a single interaction. A larger context window allows the model to consider more information at once, which matters for long documents and extended conversations.
Foundation Model: A large AI model pre-trained on huge datasets that can be adapted to many different tasks via prompting, fine-tuning or retrieval-augmented generation (RAG). Examples include GPT-4o, Claude, Gemini and Llama. Called foundation models because they serve as the base for a wide range of applications built on top of them.
Fine-Tuning: Taking a pre-trained AI model and further training it on a smaller, domain-specific dataset to improve performance on particular tasks. Fine-tuning lets organisations adapt general-purpose models to their own terminology, processes and industry context without building from scratch.
Generative AI: AI systems that can create new content (text, images, audio, video and code) by learning patterns from existing data. Generative AI sits behind tools like ChatGPT, Claude, Midjourney and DALL-E. A subset of AI, not a synonym for it.
Grounding: Anchoring an AI system’s outputs in real-world facts, knowledge or data. Grounding improves accuracy and reduces hallucinations by connecting AI responses to verified information rather than relying solely on training data.
Hallucination: When a generative AI model confidently produces output that sounds plausible but is factually incorrect, fabricated or inconsistent with its source material. A well-known limitation of LLMs. Ways to reduce the risk include RAG architectures, human review and rigorous evaluation. Non-negotiable to manage in regulated industries where accuracy matters.See: RAG (Retrieval-Augmented Generation)→
Knowledge Graph: A data structure that connects information in a web of relationships, allowing AI systems to work through and understand complex datasets. Used in search, recommendation and enterprise knowledge management.
Large Language Model (LLM): A type of AI model trained on huge quantities of text data that can understand, generate, summarise and reason about natural language. Powers tools like ChatGPT, Claude, Gemini and Copilot. The engine behind most modern AI assistants, chatbots and content tools.
Latency: The time delay between when an AI system receives an input and generates its output. Low latency matters most for real-time applications like customer service chatbots and voice assistants.
Model Card: Documentation describing an AI model: its purpose, training data, known limitations, risks, performance metrics and intended use cases. A transparency tool that guides organisations in assessing whether a model is fit for their context.
Multimodal AI: AI systems that can process and generate multiple types of data (text, images, audio and video) within a single model. GPT-4o and Gemini are examples of multimodal models.
Open-Weight vs Closed-Weight Models: Open-weight models allow an organisation to run the AI locally or in a private environment (for example Meta’s Llama). Closed-weight models are accessed only via APIs controlled by the provider (for example GPT-4o and Claude). Each carries a different governance, privacy and risk profile. Open-weight models give more control over data but require more technical capability to deploy safely.
Prompt: The input (question, instruction or context) that a user provides to an AI system to guide its output. The quality and clarity of the prompt directly affects the quality of the AI’s response.
Prompt Engineering: Designing and refining prompts to draw accurate, useful and consistent outputs from AI models. Effective prompt engineering involves clear task specification, relevant context, examples, constraints and iterative testing. An increasingly useful business skill for any role that uses AI tools.
Prompt Injection: An adversarial technique where malicious instructions are embedded in user input or retrieved content, attempting to override the AI system’s instructions and produce unintended outputs. A main security risk for AI systems that process external data.
RAG (Retrieval-Augmented Generation): An AI architecture that improves LLM responses by first retrieving relevant information from an external, verified knowledge base, then using that information to generate grounded, accurate answers. Reduces hallucinations and lets AI systems work with current, organisation-specific data without retraining the model.
System Prompt: Instructions provided to an AI model before the user interaction begins, setting its role, tone, constraints and context. System prompts are a main tool for customising AI behaviour in business applications.
Synthetic Data: Artificially generated data, produced by AI models rather than collected from real-world events or individuals. Increasingly used to expand training datasets, protect privacy in regulated industries (such as healthcare) and test AI systems without exposing sensitive information.
Token / Tokenisation: The basic units of text that large language models process. A token is approximately three-quarters of a word in English. Tokenisation is the process of breaking text into tokens for model input. Token counts determine context window limits and directly affect compute costs.
Vector Database: A specialised database designed to store, index and rapidly search high-dimensional vector embeddings. Powers semantic search, recommendation systems and RAG-based AI applications.

Section 03

Practical AI and Automation

Applying AI to real business problems. The tools, patterns and skills that make AI work in practice.

AI Agent: Software that uses a large language model or other AI techniques to perceive its environment, make decisions and take actions toward a defined goal. Unlike simple chatbots, agents can use tools, access databases, call APIs and chain multiple reasoning steps together autonomously.
Agentic AI: AI systems designed to autonomously pursue complex goals and workflows with limited direct human supervision. Where a chatbot responds to a single prompt, an agentic AI can plan a series of steps, call external tools, adapt based on results and complete multi-stage tasks independently.
Agentic Workflows: A design pattern in which AI systems autonomously plan, execute and adapt multi-step tasks with minimal human intervention. The AI agent breaks a goal into sub-tasks, calls external tools or APIs, evaluates intermediate results and iterates until the objective is met.
AI Literacy: The skills, knowledge and understanding that allow people to make informed decisions about AI systems, including awareness of the opportunities, risks and social dimensions of AI. Under the EU AI Act, AI literacy is a legal requirement for organisations deploying AI in the EU. In Australia it is increasingly expected by regulators and employers. Government staff require it from 15 June 2026.See: Tracy’s definition in Section 08→
AI Readiness: A measure of how prepared an organisation is to successfully adopt and scale AI. AI readiness spans data quality, technology infrastructure, talent and skills, governance, leadership commitment and organisational culture.See: The AI Readiness Check→
AI Strategy: A structured plan that defines how an organisation will adopt AI to achieve its business objectives. A sound AI strategy covers use case prioritisation, data foundations, technology choices, talent development, governance and a phased plan with measurable outcomes.
AI Use Policy: An internal policy document that sets boundaries, roles and responsibilities for AI tools. Defines acceptable use, prohibited use, data handling requirements and incident escalation procedures. Every organisation deploying AI should have one before staff start using AI tools at scale.
AI Washing: Overstating or misrepresenting the AI capabilities of a product or service. In Australia, AI washing is enforceable under the Australian Consumer Law (ACL) as misleading or deceptive conduct. A real risk for vendors and for organisations using AI-powered software in their own marketing claims.
AutoML (Automated Machine Learning): Tools and techniques that automate the process of building machine learning models, including feature selection, model choice and tuning. AutoML lowers the barrier to entry by reducing the need for deep data science expertise.
Conversational AI: A subfield of AI focused on systems that can understand and generate human-like language in a back-and-forth conversation. Powers chatbots, voice assistants and virtual agents.
Drift (Data / Model Drift): Performance degradation that occurs when real-world data changes away from what the model was trained on. Detected via monitoring, evaluations and periodic re-training. A main reason why AI systems need ongoing oversight, not just one-time deployment.
Human-in-the-Loop (HITL): A design pattern where humans review, approve or provide feedback on AI outputs, particularly for higher-risk tasks. HITL keeps human judgement central to decisions that matter, rather than allowing AI to operate without oversight.
Hyperautomation: Using a combination of AI, robotic process automation (RPA) and other advanced technologies to automate as many business processes as possible. Goes beyond simple task automation to orchestrate complex, end-to-end workflows.
Intelligent Automation: The combination of AI with traditional automation technologies to create systems that can not only execute predefined tasks but also learn and make decisions, with human oversight. Broader than RPA alone.
Model Monitoring: Operational tracking of an AI model’s quality, drift, latency, cost and safety incidents once it is in production. AI systems require ongoing oversight, not just one-time deployment.
Robotic Process Automation (RPA): Software that automates repetitive, rule-based tasks by mimicking human interactions with digital systems, such as copying data between applications. RPA handles structured, predictable tasks. AI extends this to unstructured data and decision-making.
Shadow AI: The use of unapproved AI tools by employees without organisational knowledge or oversight. A serious governance risk. Reduce it through clear policy, approved tool lists, staff training and monitoring.See: AI Use Policy→
Shadow IT: The broader parent category of Shadow AI: any technology used by employees without official IT or management approval. As AI tools proliferate, Shadow IT has become a primary entry point for organisational AI risk.

Section 04

Data and Privacy

Responsible data handling is the foundation of trustworthy AI. The concepts every data-handling organisation needs to understand.

Algorithmic Bias: Systematic errors in AI outputs that favour or disadvantage groups, for example by gender, ethnicity, age or location. Typically caused by biased training data or flawed model design. Manage it via representative data, fairness testing, documentation and human oversight.
Annotation: Labelling data with additional information so machine learning algorithms can learn from it. For example, labelling images as cat or dog so a model can learn to distinguish between them.
Data Governance: The policies, processes, standards and metrics that keep data managed as a strategic asset. Covers data quality, access control, privacy, lineage, classification and lifecycle management. Without data governance, AI projects cannot move safely from prototype to production.
Data Residency: The physical location where data is stored. Often overlaps with data sovereignty but is technically distinct. Residency refers to where data physically sits. Sovereignty refers to which laws apply.
Data Sovereignty: The principle that data is subject to the laws and governance structures of the nation in which it is collected and stored. Particularly relevant for Australian organisations using cloud-based AI services, where data may be processed or stored offshore.See: Data Sovereignty in practice, Section 08→
De-identification / Anonymisation: Techniques to reduce the risk that individuals can be identified from a dataset. Under Australia’s Privacy Act, de-identification is the standard term. Truly irreversible anonymisation is technically difficult in practice, especially with AI-driven re-identification capabilities.
Logging and Auditability: Recording inputs, outputs, decisions and system changes so that AI incidents can be investigated, controls verified and compliance demonstrated. Necessary for governance in regulated industries.
Personal Information (Australia): Information or an opinion about an identifiable individual, as defined in the Privacy Act 1988 (Cth). Includes obvious identifiers and any information that could reasonably identify someone. Any AI system that collects or processes personal information must comply with the Australian Privacy Principles.Source: OAIC
Privacy Impact Assessment (PIA): An analysis of privacy impacts and safeguards for projects that handle personal information. Often required, and always worth doing, for AI implementations in Australia. Should be conducted before deploying AI systems that process personal data.
Provenance / Watermarking: Markers embedded in or attached to content that indicate how it was created (for example AI-generated) or where it came from. The C2PA standard lets organisations attach machine-readable provenance metadata. Supports trust, authenticity and content moderation at scale.
Sensitive Information (Australia): A specific category of personal information under the Privacy Act 1988 (Cth), including health information, biometrics, racial or ethnic origin, political opinions, religious beliefs, sexual orientation and criminal records. Subject to higher collection and handling thresholds, generally requiring explicit consent before collection.

Section 05

AI Governance and Risk

The systems, policies and processes that keep AI safe, ethical and accountable. For boards, leaders and anyone responsible for AI decisions in their organisation.

Accountability (AI): The principle that organisations and individuals are responsible for the decisions made by AI systems they develop, deploy or use. Accountability requires clear ownership, audit trails and mechanisms for affected parties to seek redress. One of the eight principles in Australia’s AI Ethics Framework.
AI Ethics Framework (Australia): Australia’s voluntary framework, published by the Department of Industry, Science and Resources, outlining eight principles for AI: human, societal and environmental wellbeing; human-centred values; fairness; privacy protection and security; reliability and safety; transparency and explainability; contestability; and accountability.Source: DISR
AI Governance: The practical implementation of policies, processes, organisational structures and training so AI systems are developed and used in line with ethical principles, regulations and organisational values. AI governance is where good intentions become real. It turns principles into practice.See: The GIST Framework→
AI Incident: Any harmful, biased, unsafe or unexpected behaviour from an AI system, such as discriminatory outputs, privacy breaches or safety risks. Organisations should log and review incidents and have escalation and reporting processes in place.
AI Inventory (System Register): A catalogue of all AI systems and use cases across an organisation, including owners, purpose, data sources, risks and current status. Supports accountability, vendor oversight and audits. Required for Australian Government agencies under the December 2025 AI policy update.
AI Risk Assessment: A structured process to identify, analyse and reduce risks for AI systems and use cases. Often mapped to ISO/IEC 42001 (management systems) or ISO/IEC 23894 (risk guidance).
AI Risk Appetite: The level and types of risk an organisation is willing to accept in AI adoption. Guides the strength of controls, oversight requirements and the pace of AI rollout. Should be documented and approved at board or executive level.
AI Transparency Statement: A public disclosure of how an organisation uses AI, including types of systems, use cases, data handling and safeguards. Required for Australian Government agencies under the December 2025 Responsible AI Policy (v2.0). Increasingly expected in government procurement and by enterprise clients.
Algorithmic Impact Assessment (AIA): An assessment of the potential impacts of an AI system on fairness, safety and human rights. In Australia, often paired with a Privacy Impact Assessment (PIA) where personal information is involved.
Automated Decision-Making (ADM): Decisions made or substantially assisted by a computer program using personal information, without meaningful human input or with a human who relies heavily on AI output. From 10 December 2026, Privacy Act obligations require APP entities to disclose in their privacy policy what personal data is used in ADM, which decisions are made solely by a computer program and which are substantially assisted by AI. Applies even where a human is technically involved, if the decision relies heavily on AI output. Non-compliance can attract penalties over $50,000 per contravention.
Chief AI Officer (CAIO): An executive role responsible for AI strategy, governance and risk across an organisation. The Australian Government’s National AI Plan (December 2025) requires every government agency to appoint a CAIO. The role is increasingly appearing in large corporates and regulated sectors.
Contestability: The ability of individuals and organisations to challenge, question or seek review of decisions made by AI systems. One of the eight principles in Australia’s AI Ethics Framework.
Evals (Evaluation): Systematic tests for the quality, safety and resilience of AI systems, covering accuracy, bias, jailbreak resistance and other criteria. Should be repeatable and proportionate to the risk level of the system.
Explainability (XAI): The ability to understand or describe how an AI system produced a particular output. Supports accountability, trust and compliance.
Fairness (AI): The principle that AI systems should treat individuals and groups equitably, without discrimination or unjust bias. Fairness is context-dependent. What counts as fair may differ across use cases, communities and legal settings.
Human-Centred AI: An approach to AI design and deployment that keeps human needs, values and wellbeing at the centre. Prioritises usability, agency, dignity and the growth of human capability rather than its replacement.
Interpretability: How inherently understandable an AI model is, based on its architecture and logic. A model can be highly accurate but not interpretable. That tension matters most in high-stakes contexts such as credit decisions, hiring and healthcare.
ISO/IEC 42001: The international standard for AI management systems, covering policy, risk, controls, monitoring and continual improvement. The benchmark for enterprise AI governance maturity. Practical for organisations at any stage of AI adoption, including SMEs.
ISO/IEC 23894: International guidance for managing AI risks across the full system lifecycle. Complements ISO/IEC 42001 and local regulatory requirements.
Jailbreak: An adversarial prompt or technique designed to bypass an AI system’s safety controls, causing it to produce outputs it was instructed to avoid. A main security risk in enterprise AI deployments.
Responsible AI: An approach to developing and deploying AI systems that prioritises fairness, transparency, accountability, privacy and safety. In Australia, anchored by the AI Ethics Framework, privacy law and recognised international standards.
Safety Case (AI): A documented argument and evidence that an AI system is acceptably safe for its intended use. More common in high-risk sectors such as defence, healthcare and critical infrastructure.
Transparency (AI): Clarity about how an AI system works, what data it uses and how decisions are made. Supports accountability, user trust and regulator confidence. One of the eight principles in Australia’s AI Ethics Framework.
Vendor Due Diligence (AI): Assessing third-party AI tools and providers for security, privacy, reliability and compliance before adopting them. In Australia, accountability does not transfer to the vendor. If a third-party AI tool causes a discriminatory or harmful outcome, it is still your organisation’s responsibility.

Section 06

Australian AI Regulation

Australia regulates AI through existing laws rather than a standalone AI Act. That does not mean AI is unregulated. These are the frameworks, laws and regulators every Australian organisation needs to understand.

AISI (Australian AI Safety Institute): Established in November 2025 and operational from early 2026, the Australian AI Safety Institute is a government hub for monitoring, testing and analysing advanced AI capabilities, risks and harms. Funded at $29.9 million, it operates as a whole-of-government resource for AI evaluation and safety guidance.
Anti-Discrimination Law and AI: Australian federal and state anti-discrimination laws continue to apply to AI-driven decisions. Employers and service providers remain liable for discriminatory outcomes from automated systems, even where there was no discriminatory intent.
APRA CPS 234 (Information Security): Prudential Standard CPS 234, issued by the Australian Prudential Regulation Authority (APRA), requires APRA-regulated entities (banks, insurers and superannuation funds) to maintain information security capabilities proportionate to the threats they face, including those from AI and cloud platforms.
APRA CPS 230 (Operational Risk): APRA’s operational risk standard, effective 1 July 2025, requires regulated entities to manage material service provider risks, including AI vendors and cloud providers, with clear accountability and exit strategies.
Australian Consumer Law (ACL) and AI: Australia’s consumer protection law applies fully to AI products and services. Organisations must not make false or misleading claims about AI capabilities (AI washing). The ACCC has flagged AI-related harms as a priority enforcement area.
Australian Privacy Principles (APPs): The thirteen principles under the Privacy Act 1988 (Cth) that govern the handling of personal information by most Australian organisations and government agencies. Any AI system that processes personal data must comply with the APPs, covering collection, use, disclosure, quality, security and access.Source: OAIC
Automated Decision-Making: Privacy Act obligations: From 10 December 2026, APP entities using ADM software that could significantly affect individuals’ rights or interests must update their privacy policies to disclose what data is used, which decisions are fully automated and which are substantially assisted by AI. Penalties for non-compliance can exceed $50,000 per contravention.
Chief AI Officer: Government requirement: The National AI Plan (December 2025) requires every Australian Government agency to appoint a Chief AI Officer to lead AI governance and strategy. The role is now also emerging in large Australian corporates and regulated sectors.
Consumer Data Right (CDR): An Australian framework for data portability in designated sectors (banking, energy and telecommunications). Relevant when AI systems use consumer data that may be shared or accessed under CDR rules. CDR data is subject to strict accreditation and handling requirements.
Copyright and AI (Australia): Australia has no training data carve-out for AI. Organisations must have the legal right to use data for AI training. Any AI-generated content that draws from third-party material may involve copyright risk. Under active policy review as of 2026.
Digital ID Act 2024: Australia’s Digital ID Act commenced in December 2024, establishing a framework for a federated digital identity system. Private sector rollout is expected to progress through 2026 and 2027. Relevant to AI systems that verify identity or process identity-linked data.
GovAI: The Australian Government’s secure AI hosting environment for government agencies, providing a platform for deploying AI tools within government-approved data boundaries. Relevant for technology vendors and consultants working with the public sector.
IRAP (Information Security Registered Assessors Program): An Australian Signals Directorate (ASD) initiative providing a framework for assessing the security of government systems against the Australian Government Information Security Manual (ISM). Required for cloud and AI services used by Australian government agencies.
National AI Centre (NAIC): An Australian Government initiative that supports AI adoption across industry, particularly for SMEs and non-profits. Offers practical resources, programs and connections for building AI capability.
National AI Plan (December 2025): Australia’s whole-of-nation AI plan released in December 2025 by the Department of Industry, Science and Resources. Built around three goals: capture the AI opportunity; spread the benefits across the economy and community; keep Australians safe. Confirmed Australia’s standards-led approach to AI regulation, established the AISI and required a CAIO in every agency.
NDB Scheme (Notifiable Data Breaches): Australia’s scheme under the Privacy Act 1988 (Cth) requiring organisations to notify the OAIC and affected individuals when an eligible data breach is likely to cause serious harm. Applies to AI systems that process personal data. The OAIC is conducting its first compliance sweep in 2026.Source: OAIC
NSW Work Health and Safety Amendment (Digital Work Systems): New South Wales legislation governing AI-powered workplace monitoring, including performance tracking and surveillance systems. Employers must manage worker protections, and union representatives can inspect AI systems used in the workplace. Points to the direction of state-level AI workplace regulation across Australia.
OAIC (Office of the Australian Information Commissioner): Australia’s independent regulator for privacy and information access. Oversees the Privacy Act 1988 and the Notifiable Data Breaches scheme. In 2026, the OAIC is conducting its first targeted compliance sweep of privacy policies, including for ADM and AI-related obligations.Source: oaic.gov.au
Policy for the Responsible Use of AI in Government (v2.0): The updated whole-of-government AI policy (December 2025), mandatory for all non-corporate Commonwealth entities. Requires AI use case registers, risk and impact assessments, AI Transparency Statements, mandatory staff AI literacy training and a Chief AI Officer. The AI literacy requirement commenced 15 June 2026. Remaining requirements take effect December 2026.
Privacy Act 1988 (Cth): Australia’s primary federal privacy legislation, governing how organisations and government agencies handle personal information. Under active reform, with the Privacy and Other Legislation Amendment Act 2024 introducing ADM obligations (from 10 December 2026), new consent standards and strengthened individual rights.Source: OAIC
SOCI Act (Security of Critical Infrastructure Act 2018): Federal legislation that applies to AI systems in critical infrastructure sectors, including energy, water, communications, financial systems, data storage and defence. Imposes risk management program obligations on asset owners, including for digital and AI systems.
TGA and AI (Therapeutic Goods Administration): The TGA regulates AI tools that function as Software as a Medical Device (SaMD), including diagnostic algorithms, clinical decision support and AI health apps. Any AI tool used in a healthcare context that influences clinical decisions may require TGA assessment.
Voluntary AI Safety Standard: Released by the Australian Government in September 2024, this standard provides 10 guardrails for safe AI deployment. It is the current baseline most Australian organisations are measured against. The guardrails cover accountability, transparency, risk assessment, data governance, human oversight and contestability.Source: DISR

Section 07

Global and Emerging Policy

International frameworks shaping the direction of AI governance globally and what Australian organisations need to watch.

AI Safety Summits (Global): A series of international government-level summits on AI safety convened since 2023, including Bletchley Park (2023), Seoul (2024), Paris (2025) and India (2026). These summits shape international norms, testing frameworks and inter-government commitments on AI risk.
C2PA (Coalition for Content Provenance and Authenticity): An open standard for attaching provenance metadata to digital content. Lets platforms and publishers signal whether content was AI-generated or edited, supporting authenticity claims and trust in digital media. Backed by Adobe, Microsoft, Google and others.Source: c2pa.org
EU AI Act: The world’s first complete legal framework for AI, in force since August 2024 and fully applicable from 2 August 2026. Classifies AI applications into four risk categories (unacceptable, high, limited and minimal) with obligations scaling accordingly. Requires AI literacy for all staff using covered AI systems. Relevant to any Australian organisation doing business in the EU. Highly influential on emerging Australian AI policy.Source: European Commission
High-Risk AI (EU AI Act): Under the EU AI Act, AI systems used in biometrics, critical infrastructure, education, employment, migration and law enforcement are classified as high-risk. Subject to strict obligations around transparency, human oversight, accuracy and safety testing before deployment.

Section 08

Tracy’s Framework Glossary

The language of the Human-led. AI-leveraged. approach. These definitions reflect the positioning and delivery frameworks used across Tracy Sheen and The Digital Guide.

Human-led AI: An approach to AI adoption where human judgement, values and oversight stay central. AI is a tool that serves human goals, never an autonomous actor that replaces human decision-making. The human sets the direction, applies the context and takes responsibility for outcomes.See: How Tracy works with AI→
AI-leveraged: The strategic use of AI to multiply human capacity, speed and impact. AI-leveraged organisations integrate AI thoughtfully into their workflows to build on what their people do best.
The Augmented Workforce: A workforce where humans and AI systems work in partnership, each doing what they do best. Humans bring empathy, ethics, creativity and strategic judgement. AI brings speed, pattern recognition and scale. The goal is never replacement. The goal is people doing more of the work only people can do.See: The Augmented Workforce→
GIST Framework: Tracy Sheen’s signature corporate AI methodology. Guardrails. Intent. Strategy. Practical Training. A structured, human-led approach that starts with guardrails to protect the organisation, builds a strategic plan from the collective intelligence already in the room and rolls out practical training that sticks. Used in consulting, board work and corporate engagements.See: The GIST Framework→
HUMAN Framework: Tracy Sheen’s accountability map for AI governance. Five levels in a stack: Habit (the individual), Unit (the team), Macro (the organisation), Audience (clients and suppliers) and Neighbourhood (community). Each level holds itself to the same four GIST questions, so governance holds from the boardroom to the frontline.See: The HUMAN Framework→
REIMAGINE Framework: Tracy Sheen’s nine-step framework for managing the organisational change that comes with AI adoption. From the book AI & U: Reimagine Business (Australian Business Book Awards finalist 2025). The nine steps: Relationships, Evaluate, Identify, Map, Activate, Gather, Iterate, Narrow and Evolve.See: The REIMAGINE Framework→
Governance (AI): The systems, policies, processes and accountabilities that keep AI use safe, ethical and in line with organisational values and legal obligations. The rules of the road for AI in your organisation. One of the three delivery pillars across Tracy Sheen’s work.
Capability (AI): The practical skills, knowledge and confidence to use AI tools well, both at an individual level (prompting, tool use, critical evaluation of outputs) and organisationally (workflow design, upskilling, embedding new habits). One of the three delivery pillars.
Practical AI: AI adoption that starts with real business problems, not technology for its own sake. Practical AI means identifying where AI can save time, reduce errors or open up new possibilities, then implementing it safely and sustainably. One of the three delivery pillars.
AI Literacy (Tracy’s definition): The practical ability to understand, critically evaluate and confidently use AI tools in your work, without compromising your privacy, ethics or data. AI literacy is not about becoming a data scientist. It is about becoming an informed, capable user.
Data Sovereignty (in practice): Know where your data goes when you use an AI tool, which jurisdiction it falls under and whether you have the contractual rights to request deletion or access. The legal principle matters. The practical habit matters more.
Ethical AI: AI developed and used in ways that are fair, transparent, accountable and aligned with human rights and social values. Ethical AI is a practice, not a destination. It takes ongoing reflection, governance and improvement.

BEYOND DEFINITIONS

Knowing the words is the start. Using them well is the work.

Tracy walks boards, leadership teams and whole organisations through what these terms mean for their people, their risk and their Monday morning.

Get in touch →Book Tracy to speak →

Tracy Sheen in a one-to-one advisory conversation

Last updated July 2026. Definitions are provided for education. For legal or compliance advice specific to your organisation, consult a qualified professional.