AI marketing and privacy law in NZ and Australia: what your provider must do with your data.

01Is it safe to give an AI marketing provider your customer data?

It is safe when the provider takes only the data a job needs, uses AI tools that do not retain or train on it, works under a written contract that limits its use, and can tell you where it is processed. Under New Zealand and Australian privacy law, your business stays responsible for customer information it hands over.

Hiring a provider does not move the legal duty to them. The New Zealand Privacy Commissioner's guidance on cloud services puts it plainly: in most cases you remain responsible for personal information you put with a service provider. Australia reaches a similar place by a different route, covered in the offshore section below.

This guide is general information about the law as published by the regulators, and it is not legal advice. For a contract of real commercial value, or for sensitive data such as health information, have a privacy lawyer read it. For the questions to ask first, the guide to choosing an AI marketing provider covers fit, scope and price.

02Which privacy laws apply in New Zealand and Australia?

In New Zealand, the Privacy Act 2020 applies to every business that handles personal information, whatever its size, and spam is governed by the Unsolicited Electronic Messages Act 2007. In Australia, the Privacy Act 1988 and its 13 Australian Privacy Principles cover most businesses with turnover above AUD 3 million, and the Spam Act 2003 covers everyone.

New Zealand's Act is built on information privacy principles. The Privacy Commissioner lists 14 principles, the original 13 plus IPP 3A, in force from 1 May 2026¹. The ones that matter most for marketing are IPP 5 on storage and security, IPP 10 on limits to use, IPP 11 on disclosure and IPP 12 on disclosure outside New Zealand. IPP 3A matters if your provider enriches records from other sources: the Commissioner's September 2025 release explains that IPP 3A requires reasonable steps to tell people when their information was collected from someone other than them².

Australia's threshold changes who is covered. The OAIC defines a small business as one with annual turnover of AUD 3 million or less, and lists businesses covered regardless, including health service providers and businesses that trade in personal information³. A marketing provider that buys or sells contact lists can fall inside the Act even when it is small. Your provider's own status matters as much as yours.

03What do the regulators expect when AI touches personal information?

Both regulators have published guidance on AI tools. New Zealand's Privacy Commissioner expects leadership approval, a privacy impact assessment, transparency and human review before acting on outputs. Australia's OAIC recommends keeping personal information out of publicly available AI tools and checking whether a product's terms let the vendor train on your inputs.

The Commissioner's June 2023 generative AI expectations ask agencies to have senior leadership approval, conduct a privacy impact assessment, have a human review outputs before acting, and not enter personal or confidential information unless the provider has confirmed it will not be retained or disclosed⁴. The Commissioner's wider AI page states that the Privacy Act applies to everyone using AI tools in New Zealand, and advises that if in doubt, you do not use AI tools to handle personal information⁵.

The OAIC's guidance on commercially available AI products, published October 2024 and updated January 2025, says privacy obligations apply to any personal information put into an AI system and recommends against entering personal information into publicly available AI tools⁶. The same guidance tells businesses to update privacy policies to explain their use of AI, and warns that some products carry terms or settings that let the vendor collect inputs for further training.

For a marketing provider, those expectations translate into a working rule: an agent that drafts, sorts or scores using customer data runs on business-grade AI accounts with training switched off, and a person reviews anything before it reaches a customer. The glossary entry on human in the loop describes that review step.

04What data does an AI marketing provider need, and what should it never need?

Most AI marketing work runs on business information, aggregated numbers and a small set of contact fields. A provider building search, content or reporting agents rarely needs individual customer records at all. Outbound and follow-up agents need names, work emails and consent status. Almost no marketing job needs payment details, government identifiers or health information.

Two habits keep the data small. Give the provider access inside your own systems, with its own login and only the permissions the job needs, rather than sending exports by email. Where an agent only needs patterns, such as which enquiries convert, strip names before the data leaves your CRM. The Commissioner's guidance names stripping identifying details as an alternative to entering personal information into a generative AI tool.

05What happens when AI vendors process your data offshore?

Most AI models run on servers outside New Zealand and Australia. In New Zealand, sending data to an overseas provider that only processes it for you is not a disclosure under IPP 12. In Australia, an overseas cloud arrangement can count as a use rather than a disclosure under APP 8, if a contract keeps you in effective control.

The New Zealand position comes from the Privacy Commissioner's IPP 12 guidance, which says using offshore cloud providers or other agents to store or process data is not treated as a disclosure, so long as the agent is not using the information for its own purposes⁷. If an AI vendor uses your inputs for its own purposes, such as training its models, the transfer falls under IPP 12. You then need reasonable grounds to believe the recipient gives comparable safeguards, usually through a contract such as the Commissioner's model clauses, or each person's authorisation after being expressly told their information may not be protected to New Zealand standards.

Australia's test for the same arrangement is more specific. The OAIC's APP 8 guidelines, updated October 2025, say giving data to an overseas cloud provider may be a use where a binding contract limits handling to those purposes, subcontractors take on the same obligations, and the business keeps effective control⁸. Where it is a disclosure, APP 8.1 requires reasonable steps to ensure the recipient does not breach the APPs, and section 16C makes your business accountable for the recipient's breaches.

In practice, ask your provider which AI vendors it uses, where each processes data, whether training on inputs is off, and whether those terms are in a contract you can see.

06How does spam law apply to AI-written email and SMS?

Spam law applies to commercial messages however they are written, so an email drafted by an AI agent follows the same rules as one written by hand. In both countries you need consent, accurate sender details and a working unsubscribe honoured within five working days. Responsibility stays with the business whose marketing it is.

New Zealand's Department of Internal Affairs sets out three steps: express, inferred or deemed consent, clear identification of the business responsible, and a functioning unsubscribe honoured within five working days⁹. DIA warns that failure to comply could mean a fine of up to NZD 500,000¹⁰.

ACMA's guidance on the Spam Act says you must have consent before sending, that this still applies when a third party sends messages for you, and that you cannot send an electronic message asking for consent¹¹. Mixed messages count. ACMA penalised Lululemon AUD 702,900 in March 2026 for more than 370,000 order and delivery emails that carried marketing content without an unsubscribe¹². An AI agent that adds a product suggestion to a service email has turned it into a commercial message.

The guide to outbound sales automation covers how to build consent and unsubscribe handling into an outbound agent before it sends anything.

07Which contract clauses should you ask your AI marketing provider for?

Ask for a data processing agreement, or equivalent clauses in the main contract. A data processing agreement is a contract that sets out what a service provider may do with personal information it handles for you. It should cover purpose, training, location, subcontractors, security, breach notice, deletion and ownership, in plain words you can check.

• Purpose limit: the provider uses your data only to deliver the agreed scope, and for no purpose of its own.
• No training: your data and outputs are not used to train the provider's or any vendor's models, with business-grade settings confirmed.
• Vendor list: every AI model provider and subcontractor that touches personal information, the country where each processes it, and notice before any change.
• Security: access through named logins in your systems, least-privilege permissions, and no customer data in personal accounts or public AI tools.
• Breach notice: the provider tells you within a set number of hours of finding a breach, so you can meet your own reporting duties.
• Return and deletion: at the end of the work, your data is returned or deleted, with written confirmation.
• Ownership: accounts, lists, prompts, agent instructions and outputs belong to your business.
• Spam compliance: consent records and unsubscribe handling are kept, and nothing is sent to a list without a recorded consent basis.

The breach clause matters because the clock runs on you. New Zealand's Privacy Commissioner expects a notifiable privacy breach to be reported as soon as practicable, and within 72 hours of becoming aware it is notifiable¹³. Australia's Notifiable Data Breaches scheme gives a covered business 30 days to assess a suspected breach and requires notice as soon as practicable once it believes an eligible breach has occurred¹⁴.

08A privacy checklist for hiring an AI marketing provider

Use this checklist before you share any customer data. Each line names the principle it rests on in New Zealand and Australia, so you can find the regulator's guidance if a provider pushes back. A provider that answers every line in writing, quickly, has done this before.

For the country-specific side of hiring, the pages on AI marketing services for New Zealand businesses and AI marketing for Australian teams cover what changes on each side of the Tasman.

09What is changing in Australian privacy law?

Australia is part way through a two-stage reform. The first stage passed in late 2024: a statutory tort for serious invasions of privacy started in June 2025, and automated decision transparency starts on 10 December 2026. A second stage was released as an exposure draft in August 2026 and remains a draft.

The OAIC confirms the statutory tort for serious invasions of privacy commenced on 10 June 2025¹⁵. The OAIC's May 2026 consultation confirms that from 10 December 2026, businesses covered by the Act that use personal information in automated decisions that could affect people's rights or interests must describe the kinds of decisions and information in their privacy policies¹⁶. If an agent decides who gets an offer or what price someone sees, check whether it falls in scope.

The Attorney-General's Department opened consultation on the exposure draft Privacy Amendment (Personal Data Protection) Bill 2026 on 31 August 2026, with submissions closing 18 September, and says it remains subject to further consideration by government¹⁷. The draft proposes a controller and processor split, with an exception for processors acting on a controller's behalf, and a rewritten APP 7 that requires a simple opt-out on direct marketing. Plan on current law until it passes.

10Where Flow AI fits

Flow AI provides AI marketing services for businesses in New Zealand and Australia. Amber Lan, an AI marketing engineer in Auckland, builds the AI agents that run search, content, outbound and reporting with your team, then hands them over.

Each build starts with a written scope, which is the natural place to agree what data each agent may touch, and ends with a documented handover so your team owns and runs the system. The agent templates are public, so you can read how an agent is instructed before you share anything. Companies hiring a marketing engineer are welcome to reach Amber about roles.

11The short version

Giving customer data to an AI marketing provider is safe when the provider takes only what each job needs, keeps it out of AI tools that retain or train on it, and signs a contract that limits use, names its vendors and sets breach and deletion terms. Your business stays responsible in both countries, including for offshore processing.

Spam law applies to every AI-written message, so consent records and a working unsubscribe come before the first send. For how this fits a wider programme, the guide to AI marketing transformation for a small business sets out the order of work.

To scope an agent build together, with data rules written in from the start, talk to Amber.