What Your Staff Are Pasting Into the Chatbot

Here is the honest shape of it. Somebody in your business has pasted a customer's email thread into an AI tool to get help writing a reply. Somebody has pasted a contract in to have it summarised. Somebody has pasted a spreadsheet of names to have it tidied up. Nobody asked you, because it did not occur to them that it was the sort of thing you ask about. It felt like using a calculator.

The scale is smaller than the noise suggests, and closer than you think

The Australian Bureau of Statistics measures this on a proper sample frame, which is more than can be said for most of the numbers you will see quoted. In 2024-25, around 12 per cent of Australian businesses reported using AI in their workplace. Uptake ran at about 35 per cent for large businesses, up from 9 per cent in 2021-22; 22 per cent for medium ones, up from 3 per cent; and around 11 per cent for small and micro businesses (ABS, 2026, reference period 2024-25).

That 11 per cent is the number of small businesses that reported using AI. It is not the number where somebody used a chatbot last week. Those are different questions, and I would bet the second number is a multiple of the first, because the thing being measured is a business decision and the thing actually happening is an individual habit. The gap between them is precisely where the risk lives - nobody governs a tool the business does not believe it uses.

The ABS also found something worth noticing: small businesses that were actively innovating adopted AI at 19 per cent, close to five times the rate of those that weren't. The busy, curious end of small business is well ahead of the average. If that sounds like your place, assume the tools are already in the building.

What the regulator actually says

The Office of the Australian Information Commissioner published guidance on using commercially available AI products in October 2024, updated in January 2025. Two parts of it matter to you.

The first is a statement of scope: privacy obligations apply to any personal information put into an AI system, and to the output where the output contains personal information (OAIC, 2025). Pasting a customer's details into a chatbot is a use and possibly a disclosure of their information. It is not a private act just because it happened on your own screen.

The second is blunter, and it is the OAIC's own recommendation rather than a rule: organisations should not enter personal information, and particularly sensitive information, into publicly available generative AI tools. Not "should be careful". Should not.

Now, whether the Privacy Act binds you depends on your turnover and your industry. Businesses turning over $3 million or less sit outside it by default, though the carve-outs are wider than most owners realise: health service providers, childcare and anyone trading in personal information are in, whatever their size. Either way, treat the OAIC line as the sensible default. A customer who finds their medical history in a chatbot log will not be comforted to hear you were under the threshold.

The question is where the words go

Strip the technology away and the risk is old and familiar: information leaving the building through a channel nobody is watching. We had this problem with fax machines, then USB sticks, then personal email. The chatbot is the current one. Three things determine whether a paste is a problem.

Which account. A free consumer account and a paid business account on the same product are not the same thing, and the difference is in the terms rather than the interface. Business tiers commonly commit not to train on your inputs and give you some control over retention; consumer tiers commonly reserve the right to do the opposite. I will not tell you what any particular vendor's terms say this month, because they change. Open the terms for the tier you are on and find the two sentences about training and retention. Same interface, same logo, different contract.

What was in it. A rewrite of your own marketing blurb is nothing. A customer list, a contract with commercial terms, an employee's health details, a client's financial position - those are somebody else's information, and you are holding them on trust.

Whether anyone could tell you afterwards. This is the one owners find most uncomfortable. If I asked you today what had been pasted into an AI tool from your business in the last month, could anyone give me an answer better than a shrug?

A one-page rule beats a policy nobody reads

You do not need a governance framework. You need a page, and a conversation at a Monday morning meeting.

Name the tools you have approved, and pay for the business tier on the one or two that people actually use. The cost is trivial next to what it buys, which is a contract that says your work is not training data. This is the same shape as every other control worth having in a small business: sized for you, not for a bank, and cheap enough that the decision is not really about money.

Then write two lists, short enough that people remember them without looking.

Fine to paste:

• Your own drafts, notes and marketing copy.
• General how-do-I questions with no names in them.
• Code you own.
• Anything already published, and anything you would happily read out at a counter.

Never paste:

• Customer or client personal details.
• Health information, which the Act treats as sensitive, and which - if you provide a health service - puts you inside the Act whatever your turnover.
• Anything under a confidentiality clause, and anyone else's financial records.
• Credentials of any kind - passwords, API keys, recovery codes, the lot.

That last item deserves its own sentence. A credential pasted into a chat window is a credential you have now lost track of, and it goes straight onto the list of keys you cannot account for.

Then say the thing that makes it work: nobody is in trouble for asking, and nobody is in trouble for a paste they made before this conversation. You want people telling you what they use. The moment using AI feels like something to hide, it goes underground and you lose the only visibility you had.

The best version of this I have seen took eleven minutes at a staff meeting and lived on a laminated page next to the kettle. The worst was a fourteen-page policy at a much larger organisation that nobody under management level had read, which achieved nothing except a defensible paper trail.

Security advice written for big organisations assumes a compliance function that will absorb it. You do not have one. What you have is a short list of people who will do the sensible thing if somebody tells them what it is, and who will otherwise guess. This is worth ten minutes of telling.

What has gone into a chatbot from your business this month? If your honest answer is that you have no idea, that is not a failure. It is just the first thing worth fixing.

---

Sources

• Australian Bureau of Statistics, Business adoption of Artificial Intelligence accelerates in 2024-25 - retrieved 5 August 2026, https://www.abs.gov.au/media-centre/media-releases/business-adoption-artificial-intelligence-accelerates-2024-25
• Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products - retrieved 5 August 2026, https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products
• Office of the Australian Information Commissioner, Rights and responsibilities under the Privacy Act - retrieved 5 August 2026, https://www.oaic.gov.au/privacy/privacy-legislation/the-privacy-act/rights-and-responsibilities
• Red Bridge Cyber, perspective: Security Advice Sized for Someone Else
• Red Bridge Cyber, perspective: Who Still Has the Keys?
• Red Bridge Cyber, guide: Small business cyber security explained