Keeping client data private when you use AI
The Privacy Commissioner's office says the Privacy Act applies to everyone using AI tools in New Zealand, and that the privacy principles cover what you type in. Its advice is that if you're in doubt, don't use AI tools on personal information. Business.govt.nz treats typing, pasting and uploading as disclosure. So the simplest rule is to treat a prompt the way you'd treat an email to an outside company, because in many cases that's where it's going.

Take personal details out first
Start with whether you need the personal details at all. Most of what people use AI for, like summarising a document or drafting a letter, works just as well with the names, addresses and account numbers taken out.
Check the terms of your plan
Next, look at the plan you're on, because the terms depend on it. OpenAI says it doesn't train by default on ChatGPT Business, Enterprise, Edu or API data, but free, Plus and Pro content may be used unless the user opts out. Anthropic says it doesn't train by default on its commercial products unless the customer allows it. Google says Workspace doesn't use customer data for training without permission. Microsoft says prompts and responses aren't used to train the models behind Microsoft 365 Copilot. These terms change, so read them yourself. We use Anthropic's AI assistant in our own work, so we have an interest in saying that plainly.
Sending personal information overseas
Not training on your data doesn't make it stay in New Zealand. Microsoft, for example, says customers outside the EU may have queries processed in the US, EU or other regions. Principle 12 of the Privacy Act limits sending personal information overseas to listed grounds, such as the person's authorisation after being told the risk, or a reasonable belief that the recipient has comparable safeguards. The Commissioner's older guidance on cloud services says you stay responsible for what you put there. How that applies to a particular AI vendor is a question for your lawyer.
Privacy Principle 3A
There's also a newer rule. Since 1 May 2026, Principle 3A has generally required you to take reasonable steps to tell people when you collect their personal information from another source. The Commissioner's pages we read don't apply it directly to AI tools. Our reading is that it's worth thinking about if an AI tool is gathering information about people for you.
What the Privacy Commissioner expects
If you're bringing in generative AI properly, the Commissioner expects senior approval, a check that the tool is necessary and proportionate, a privacy impact assessment, transparency, engagement with Māori, and human review before acting on the output. For a small business, that can be one page. Write down which tools you use, what can go in them, what can't, and who checks the results.
If there's a privacy breach
If something does go wrong, the Act requires you to notify the Commissioner and affected people as soon as practicable when a breach has caused or is likely to cause serious harm. The Commissioner uses 72 hours as a guide, although it isn't a legal limit.
This is general information, not legal advice.
Sources
Sources: Office of the Privacy Commissioner, "AI", "Generative Artificial Intelligence" and "How long is 72 hours?", and IPP3A and overseas disclosure pages. Privacy Act 2020. business.govt.nz, "Safe and smart AI use", edited 1 April 2026. OpenAI, Anthropic, Google and Microsoft data use pages, read November 2026.




Comments