What is safe to paste into AI
Keep anything that names a person, or could show who they are, out of AI tools, unless you know the provider does not keep it or pass it on. That follows the Privacy Commissioner's advice, and the Privacy Act applies to New Zealand organisations of every size. Below is what is usually safe to paste, what is not, and what to check before you pick a tool, with each source linked. It is general information, not legal advice.
What the Privacy Commissioner says
The Privacy Commissioner says "Privacy law applies to the use of AI tools." The Commissioner's advice when you are not sure: "the safest approach is to avoid putting personal information into AI tools, and to make sure that is clear to everyone across your organisation." (Artificial intelligence and the Information Privacy Principles, September 2023)
So the rule I give every team is simple. If a piece of text names a person, or says enough that someone could work out who it is about, take those details out before it goes in.
Usually safe to paste
- Text about your work that does not identify anyone and is not confidential: a policy draft, a process you want to tidy up, a job ad, a product description.
- A real email or report after you have taken out names, contact details and anything else that points to a person.
- Material you have already published, such as your website.
Not safe to paste
- Names, contact details or health details of customers, patients or staff, unless you know what the tool's provider does with them. Check this before you use any free public tool. The Commissioner's office says: "If in doubt, we recommend you do not use AI tools to handle personal information." (the Commissioner's AI page)
- Confidential information. The Commissioner's statement from June 2023 says: "Do not input into a generative AI tool personal or confidential information, unless it has been explicitly confirmed that inputted information is not retained or disclosed by the tool provider." (Generative AI statement, June 2023)
Check these before you pick a tool
These checks come from the Commissioner's AI guidance, unless another link is given.
- The terms. The Commissioner's guidance gives questions like: "Will the provider store information I put?" and "Will it be used by the provider to train their AI systems?" The guidance points to tools you can run offline, or tools "where the contractual terms stop the provider from using input data for training."
- Why you collected the information. Typing someone's details into a prompt "may go beyond the purpose for which you collected that information."
- Where the data goes. If an offshore provider only stores or processes your data for you, the Commissioner says that "is not treated as a disclosure under IPP12, so long as they are not using that information for their own purposes." (IPP12 is the rule on sending personal information overseas.) Whether this applies "may depend on the details of AI tool being used", and a provider that trains its AI on what you type is likely using it for its own purposes. Either way, "you remain responsible for protecting that information."
- Security. In the Commissioner's words, "Some AI tools can leak sensitive information that increases security risks."
- A privacy impact assessment: a written check of what personal information the tool will handle, what could go wrong, and how you will prevent it. Among the things the Commissioner expects is to "Conduct a privacy impact assessment before using these tools."
- Faces, voices and fingerprints. If a tool uses any of these to recognise people or learn about them, check the Biometric Processing Privacy Code 2025 first. It came into force on 3 November 2025, and the grace period for those already using biometrics ended on 3 August 2026.
A person checks the output
AI writes with confidence even when it is wrong. The Commissioner puts it this way: "Generative AI tools produce confident errors of fact and logic, so you cannot rely on their apparent performance or self-declared capabilities as evidence of accuracy." The Commissioner expects "human review prior to acting on AI outputs". (AI guidance) In practice, a person on your team reads every AI draft before it leaves.
The Commissioner's AI guidance dates from September 2023, and the office says it plans "to update our guidance over time". Check the links above for the latest version.
How I set this up with a team
When I work with a team, names and identifying details come out before any text goes into an AI tool. Staff are trained on what is safe to enter and what is not. A person checks the work before it leaves. We write down which tools are used and where the data is stored, and you keep that page.
That is part of how AI consulting works with me: we cut the paperwork first, then put AI on what is left.
Where to start
Start with the scan. 15 to 20 minutes, one person. You get a one-page map of which task to fix first.