Is my business data safe if I use AI tools?

The distinction that matters most

The same brand of AI tool can behave completely differently depending on which tier you are on. Free and consumer subscriptions commonly reserve the right to retain what you type and use it to improve the service. Business, team and enterprise tiers generally commit in the contract not to train on your inputs, and expose controls over how long data is kept.

This matters because the staff member who starts using AI at work is usually on a personal account they already had. That is not misconduct and it is not a technology failure — it is the predictable result of nobody having said which account to use. Stating that is the single cheapest control available to a small business.

What the Privacy Act makes you responsible for

If you handle personal information and the Australian Privacy Principles apply to your business, handing that information to an AI vendor does not transfer your obligations. You remain accountable for the disclosure. APP 8 in particular deals with sending personal information overseas, and most major AI providers process outside Australia.

The practical consequence is that "the vendor is certified" is an answer to a different question. The vendor being secure is necessary and not sufficient; what you need is a defensible account of why that disclosure was reasonable and what you told people about it. This is general information about where the obligations sit, not legal advice — for an assessment of your own obligations, talk to a lawyer.

The four questions to ask before client data goes in

• Is my input used to train or improve your models, and is that stated in the contract rather than a blog post? A commitment in marketing copy can change without notice; one in the agreement cannot.
• How long is my data retained, and can I set that to zero? Many business tiers offer configurable or zero retention. Ask where logs sit too — retention often applies to the model and not the surrounding infrastructure.
• Which country does processing happen in, and are there Australian options? This is the APP 8 question, and the answer changes what you have to disclose.
• Who inside the vendor can see my content, and under what circumstances? Nearly every provider allows some human review for safety or abuse investigation. That is reasonable, but you should know it before promising a client otherwise.

The risk that is not about vendors at all

Most real incidents are not a provider being breached. They are an internal access problem that AI made visible. Connect an AI assistant to a shared drive and it will faithfully retrieve whatever the person asking is permitted to see — including the payroll spreadsheet in a folder that was quietly shared with everyone in 2021.

The permissions were always wrong. What changed is that finding the file no longer requires knowing it exists. Before connecting any AI tool to a document store, audit what the connecting account can actually reach, because that is precisely the blast radius.

A proportionate position for a small business

You do not need an enterprise governance programme. You need four decisions written down somewhere findable: which tools are approved, which categories of information must never be entered, that work goes through work accounts rather than personal ones, and who to tell if something goes in that should not have. One page is enough, and one page that exists beats a framework that is still being drafted.