Your staff are probably already using AI

You may not have formally introduced AI into your business. That doesn't mean it isn't being used. A staff member may ask a public chatbot to improve a customer email. Someone may upload a spreadsheet for analysis, summarise meeting notes, create a proposal or paste in a difficult complaint to help draft a response.

They may be doing it for a perfectly understandable reason: it makes the task easier. The problem is not necessarily the use of AI. The problem is that the business may not know which tools are being used, what information is entering them or how much trust is being placed in the result.

Informal AI adoption happens quickly

Most new business software requires some effort to introduce. It may need installation, training, configuration or a purchasing decision. Public AI tools have almost none of those barriers. Staff can create personal accounts and begin using them without involving the owner or IT provider.

Common examples include:

• Pasting an email into a chatbot and asking it to prepare a reply.
• Uploading a document for summarisation.
• Entering customer details to produce a report.
• Using AI meeting transcription without discussing it with attendees.
• Generating marketing material from internal business information.
• Asking AI to assess job applications.
• Using personal AI accounts for business work.
• Installing browser extensions that can access the contents of webpages.

Not all these uses carry the same risk. Drafting a heading from public information is very different from uploading a customer's financial or health information. If the business hasn't explained the distinction, staff are left to make that judgement themselves.

Start by finding out what's already happening

The first response shouldn't be an accusatory email or an immediate ban. Ask people how they are currently using AI and what they are trying to achieve. Make it clear that the purpose is to understand the activity and establish reasonable boundaries — not to catch anyone out.

Useful questions include:

• Which AI tools are you using for work?
• Are you using a personal or business account?
• What tasks do they help with?
• What information do you enter?
• Do you upload files, images, recordings or spreadsheets?
• How do you check the result?
• Has the tool ever produced something clearly wrong?
• Are there tasks you would like help with but don't currently use AI for?

This conversation may identify risks. It may also uncover useful opportunities that management hadn't considered. Someone repeatedly using AI to summarise the same kind of document is telling you something about the process. The informal workaround may be unsuitable, but the underlying need may be worth addressing properly.

The information risk is easy to underestimate

Entering information into an AI tool is still sharing information with a system outside the employee's immediate control. Depending on the product, account type, terms and settings, submitted content may be retained, reviewed, used to improve services or made available to connected services. The details vary, which is precisely why they need to be checked.

The Australian Cyber Security Centre advises small businesses to understand:

• What data the tool collects.
• Where that information is stored.
• Who owns it.
• Whether it will be used to train AI models.
• How the system's outputs will be checked.

It also recommends providing staff with AI-related security training. See the ACSC's AI guidance for small businesses.

The Office of the Australian Information Commissioner goes further for public generative AI tools. As a matter of best practice, it recommends that organisations do not enter personal information — particularly sensitive information — into publicly available generative AI tools because of the privacy risks. Read the OAIC guidance.

This doesn't mean every use of AI is unsafe. It means the business needs to understand the particular tool, information and use case rather than assuming all AI services operate in the same way.

A careless upload can become the business's data breach

Consider a staff member who receives a lengthy customer document. To save time, they upload it to a public AI tool and ask for a summary. The document contains the customer's name, contact details, financial circumstances and information about their health.

That action may feel less serious than emailing the document to the wrong person. From a privacy perspective, however, information may still have been disclosed outside the business. The OAIC says that entering personal information into an AI system may be considered a disclosure if the information becomes accessible outside the organisation and is released from its effective control.

In workplace guidance published in December 2025, the OAIC used a fictional example based on a notification made under the Notifiable Data Breaches scheme. An employee uploaded a customer's financial hardship application — including health and family information — to a publicly available AI tool to prepare a summary. The OAIC used the example to demonstrate how inappropriate workplace AI use can lead to a data breach, harm to an individual and possible regulatory consequences. Read the OAIC workplace example.

Staff must know that an accidental upload should be reported internally — not concealed, deleted from their browser history or quietly handled on their own. The business needs the opportunity to contain the incident, establish what happened, contact the provider if appropriate and assess whether further action is required.

A business account is not a complete answer

Moving staff from personal accounts to an approved business or enterprise account can provide better administrative control, but the name of the plan isn't enough. Check:

• How submitted information is handled.
• Whether it is used for model training.
• Available retention controls.
• Where information is processed or stored.
• Whether access can be removed centrally.
• What integrations and extensions have been enabled.
• Whether activity can be audited.
• What happens to business information when a staff member leaves.

Settings and contractual terms matter more than reassuring product language.

Give staff usable boundaries

A policy that simply says "use AI responsibly" gives employees very little help. A practical small-business policy should answer five things.

1. Which tools are approved?

Name the tools and account types employees may use. Explain whether personal accounts are permitted for business work.

2. What can they be used for?

Examples might include brainstorming based on public information, improving the structure of non-confidential writing, summarising approved internal material, drafting content that will be reviewed before publication, and assisting with defined administrative tasks. Use examples relevant to the work your staff actually perform.

3. What information must not be entered?

Depending on the business, this may include customer or patient information, employee information, passwords and access credentials, financial records, legal advice or privileged material, confidential commercial information, unpublished intellectual property, complete internal databases, and information supplied under confidentiality obligations. "Don't enter sensitive data" is too vague unless staff know what the business considers sensitive.

4. What must be checked?

Employees should remain responsible for work they produce with AI assistance. Specify when facts, figures, sources, calculations and customer details must be verified. Higher-consequence outputs may require approval by another person before use. Polished language can make an incorrect answer appear more credible. Presentation is not evidence of accuracy.

5. What happens if something goes wrong?

Give staff a clear reporting path. If information is uploaded accidentally, an incorrect output reaches a customer or an unapproved tool gains access to business systems, people need to know whom to tell. The policy should identify the person to contact, how quickly incidents must be reported, the details that should be preserved, what the employee should avoid doing, and who decides whether customers, advisers, insurers or regulators must be contacted. Fast reporting gives the business a better chance to prevent or reduce harm. A culture that punishes every honest mistake may encourage people to conceal the next one.

Consider a simple traffic-light guide

For a small team, a short guide may be more useful than a long policy.

The exact categories should reflect the business, its contractual obligations and any laws that apply to it.

Don't make the policy entirely restrictive

If AI can genuinely help with a task, provide an approved way to use it. Otherwise, staff may continue using whatever is easiest, quietly and through personal accounts. A policy works better when it combines boundaries with practical alternatives.

This may mean providing an approved business account, configuring privacy and retention settings, creating safe templates for common tasks, removing personal information before approved use, building a controlled workflow for a recurring need, requiring human approval before an action is completed, or deciding that a particular use is not appropriate.

The aim is visibility, not surveillance

You do not need to record every prompt an employee writes. You do need to know which tools form part of your business, what information they handle and who is responsible for their use.

Staff are likely to keep exploring AI because some of it is genuinely useful. The business has a choice between leaving that use informal or giving people the knowledge, tools and boundaries to use it appropriately.

Start with a conversation. You may discover that AI adoption has already begun.

This article provides general business information and is not legal or privacy advice. Requirements will depend on your business, industry and the information you handle.