Glossary

Cloud, security and AI, without the jargon.

Plain-English definitions of the terms used across the OneNet site, so you can read with confidence whether or not you live in this world day to day.

How to use this

Search for a term, or jump to a letter. Definitions are intentionally short, enough to make a conversation easier without becoming a textbook.

A

Adversarial attack

Any attempt to manipulate an AI system into producing the wrong answer or behaving unsafely.

Agent

An AI system that can take actions on your behalf, looking up information, calling other systems, completing multi-step tasks, not just answering questions.

AI (Artificial Intelligence)

Computer systems that perform tasks normally associated with human intelligence, recognising patterns, generating text or images, making decisions.

AI governance

The policies, decision rights and oversight that make sure AI is used responsibly and in line with your obligations.

AI readiness

How prepared your people, processes, data and technology are to adopt AI safely and get value from it.

API

Application Programming Interface. The standard way one piece of software talks to another.

B

Backup

A copy of your data kept separately, so that you can restore it if the original is lost, corrupted or attacked.

BCP (Business Continuity Plan)

The wider plan that keeps the business operating during disruption, of which IT disaster recovery is one part.

C

CAIO (Chief AI Officer)

The executive responsible for an organisation's AI strategy, governance and value, a role many businesses are now creating alongside the CIO and CISO.

CDC Auckland

The Canberra Data Centres facility in Auckland, a Tier III+ data centre where OneNet hosts its core New Zealand cloud platform.

CIO / CISO / CTO

Chief Information Officer, Chief Information Security Officer and Chief Technology Officer, the senior technology and security roles in most organisations.

CIS (Center for Internet Security)

A US-based non-profit that publishes globally recognised cybersecurity benchmarks and controls for hardening systems against common threats.

Cloud

Computing services, servers, storage, networking, software, delivered over the internet rather than from a server in your office.

Conditional access

Rules that decide whether someone can sign in based on context, who they are, what device they're on, where they are and how risky the sign-in looks.

Copilot

An AI assistant embedded inside the tools you already use, helping you draft, summarise or analyse without leaving the application.

D

Data poisoning

An attack where bad data is deliberately introduced into AI training material to make the resulting model behave incorrectly.

Data sovereignty

The principle that data is subject to the laws of the country it is stored in. Hosting in New Zealand keeps data under New Zealand law.

DLP (Data Loss Prevention)

Tools and policies that stop sensitive data, such as client records or IP, from being shared or sent where it shouldn't go.

DNS

Domain Name System. The internet's address book, turns names like onenet.co.nz into the numeric addresses computers use.

DR (Disaster Recovery)

The plan, processes and technology that get critical systems running again after a major outage or incident.

E

EDR / XDR

Endpoint Detection & Response (and the broader Extended version), security tooling that watches laptops, servers and other endpoints for malicious behaviour.

Embedding

A numerical representation of a piece of text, image or data that lets an AI compare meaning rather than just words.

EU AI Act

European Union legislation that places obligations on organisations developing or using AI, with stricter rules for higher-risk uses.

F

Fine-tuning

Adapting a general AI model to your specific domain, language or style by training it further on your own examples.

G

GDPR

The European Union's data protection law. Often used as a benchmark for handling personal information well.

Generative AI

AI that creates new content, text, images, code, audio, rather than just classifying or analysing existing data.

Guardrails

Rules and controls that keep an AI system on-topic, on-policy and away from unsafe behaviour.

H

Hallucination

When an AI produces an answer that sounds confident but is wrong or made up. Grounding AI in your real content reduces this.

Hybrid cloud

A mix of private and public cloud working together, so each workload can run where it makes the most sense.

Hyperscaler

One of the very large public cloud providers, most commonly Microsoft Azure, Amazon Web Services and Google Cloud.

I

Immutable backup

A backup that cannot be changed or deleted for a set period, a key defence against ransomware.

Inference

The act of an AI model producing an answer from a prompt, as opposed to training, which is how the model was built.

ISO/IEC 27001

An international standard for managing information security. Certification means an organisation has independently verified controls in place.

ISO/IEC 42001

An international standard specifically for managing AI responsibly across an organisation.

J

Jailbreak

A technique used to make an AI model bypass its safety controls and produce content it normally wouldn't.

L

LLM (Large Language Model)

An AI model trained on huge amounts of text that can read, write and reason in natural language. ChatGPT and Microsoft Copilot are powered by LLMs.

M

Managed desktop

A service where someone else takes responsibility for setting up, securing and managing the laptops and desktops your team uses every day.

MDR (Managed Detection & Response)

A managed service where a security team monitors your systems for threats around the clock and steps in to contain them when they appear.

MFA (Multi-Factor Authentication)

Logging in with more than just a password, for example a code from your phone, so a stolen password isn't enough on its own.

Microsoft 365 (M365)

Microsoft's cloud suite, including Outlook, Teams, Word, Excel, SharePoint and OneDrive, used by most modern workplaces.

MITRE ATLAS

A knowledge base of techniques attackers use against AI and machine learning systems, essentially the AI equivalent of MITRE ATT&CK.

Model

The trained mathematical 'brain' behind an AI system. Different models are good at different things.

Model drift

The gradual decline in an AI model's accuracy over time as the world it was trained on changes.

Model inversion / extraction

Techniques used to reconstruct the data a model was trained on, or to clone the model itself, by probing it carefully.

N

NIST AI Risk Management Framework

A framework from the US National Institute of Standards and Technology that helps organisations identify and manage risks unique to AI systems.

O

On-premisesOn-prem

IT systems running on hardware physically located in your own offices, rather than in a cloud or data centre.

Orchestration

Coordinating multiple AI models, tools and steps so they work together to complete a larger task.

OWASP Top 10 for LLMs

A community-maintained list of the most common and serious security risks for applications built on large language models.

P

Penetration test

An authorised, simulated attack against your systems to find weaknesses before a real attacker does.

Phishing

Fake emails, messages or websites designed to trick people into handing over passwords, money or sensitive information.

PII (Personally Identifiable Information)

Information that can identify a specific person, names, contact details, ID numbers, and which needs to be handled carefully under privacy law.

Private cloud

Cloud infrastructure dedicated to a single organisation, giving more control over performance, data location and security than a shared public cloud.

Prompt

The instruction or question you give an AI system to tell it what you want it to do.

Prompt injection

An attack where someone hides instructions inside content the AI reads, tricking it into ignoring its rules or leaking data.

Public cloud

Cloud infrastructure shared across many tenants and operated by a hyperscaler such as Microsoft Azure or Amazon Web Services (AWS).

R

RAG (Retrieval-Augmented Generation)

An approach where the AI looks up relevant information from your own content first, then uses it to write a more accurate answer.

Ransomware

Malicious software that locks up or encrypts your data and demands a "ransom" payment to unlock it.

Red team

A group that simulates real adversaries to test how well people, processes and technology hold up under a cyber-attack.

RFP / RFI / Tender

A formal request from a buyer asking suppliers to describe how they would meet a need (RFP = Request for Proposal, RFI = Request for Information).

ROI (Return on Investment)

What you get back from an investment compared to what you put in, used to judge whether a project is worth doing.

RPO (Recovery Point Objective)

The maximum amount of data, measured in time, you're prepared to lose in an incident, for example 'no more than 15 minutes of work'.

RTO (Recovery Time Objective)

The maximum time you're prepared to be without a system after an incident before it has to be back online.

S

SaaS / PaaS / IaaS

Different layers of cloud service: Software, Platform and Infrastructure as a Service. Each shifts more of the technical responsibility to the provider.

SD-WAN

Software-Defined WAN. A modern way of running a WAN that uses software to send traffic over the best available link, improving performance and resilience.

Shadow AI

AI tools being used inside an organisation without IT or leadership knowing, a major source of unmanaged risk.

SIEM

Security Information & Event Management. A platform that collects logs from across your environment so threats can be detected and investigated.

SLA (Service Level Agreement)

A written commitment about how a service will perform, for example uptime, response times or resolution times.

SOC (Security Operations Centre)

The team and tooling that watches for, investigates and responds to cyber threats, typically operating 24/7.

SSO (Single Sign-On)

One login that gets you into multiple applications, instead of separate usernames and passwords for each.

T

Telemetry

Operational data, logs, metrics, events, that lets you see how a system is behaving so you can monitor and improve it.

Threat model

A structured way of mapping who might attack a system, how they'd do it, and what you need to defend.

Tier III+ data centre

An industry rating for highly resilient data centres, with redundant power, cooling and networking and very little permitted downtime each year.

V

Vector database

A database designed to store embeddings and quickly find the most similar items, the engine behind much of modern AI search.

Veeam

A widely used enterprise backup and recovery platform that OneNet uses to protect client data.

VPN

Virtual Private Network. A secure, encrypted tunnel that lets users or sites connect to a network over the public internet as if they were directly attached.

W

WAN (Wide Area Network)

The network that connects multiple sites, offices, data centres, cloud, together over distance.

Workload

Any application, service or process that runs on IT infrastructure, for example a website, a database or an AI model.

Z

Zero Trust

A security approach that assumes no user, device or network is automatically trusted. Every request is verified before access is granted.

Let's talk

Ready to move faster, safer, smarter?

Book a free 30-minute strategy call. No slides. No sales pitch. Just an honest conversation about your cloud and AI roadmap.

0/250

We'll reply within one business day. Promise. This site is protected by reCAPTCHA, Google's Privacy Policy and Terms apply.

Prefer to talk now?

24/7 support
All systems operational