You're offline, showing the last version we saved.
Build with AI · Using AI safely

Client data and AI: a beginner’s guide

Most of the risk with AI isn’t the AI. It’s what gets typed into it. Here’s what happens to what you type, what’s safe to put in, and five steps to set your firm up.

For beginners · Checked 11 November 2026 · Not advice

In plain English

When you type into an AI chat, it goes to the AI company’s servers, usually overseas. Depending on your plan and settings, they may keep it for a while and may use it to improve their models.

Once it’s sent, you can’t take it back. So the safest habit is simple: client details stay out of the chat. A quick test: if you wouldn’t be happy to see it on the front page of a newspaper, don’t put it in.

What happens when you press enter

  1. You type or attach Your prompt and any file you add are sent over an encrypted connection to the AI provider.
  2. The provider processes it On its servers, often overseas, sometimes on other companies’ cloud services.
  3. It’s kept for a while Logs are kept for a period. Depending on your plan it may train models, and flagged chats can be reviewed by the provider’s staff.

What can go into an AI tool?

  • Go: Fine to use

    • Your own writing, templates and checklists with no client details
    • Public information: ATO guidance, legislation, published rates
    • Made-up examples and sample data
    • Text with client details swapped for placeholders

  • Take care: Only in an approved business tool

    • Client work with names and identifiers removed
    • Internal documents: procedures, pricing, staff guides
    • Anything your AI policy allows, with training switched off

  • Stop: Never in an AI chat

    • TFNs, bank, Medicare, licence and passport numbers
    • Client files, returns and financials with names on them
    • Health, family or other sensitive details about a client
    • Passwords, logins and API keys

Some firms approve an enterprise tool for more, such as an AI assistant that works inside the firm’s own Microsoft or Google account. That’s a firm decision with your IT provider, written into your AI policy, not one person’s call.

Does the AI learn from what I type?

Free and personal plans
Often yes, unless you switch it off. Look for a setting like “Improve the model” or “Help improve” in the privacy or data controls settings.
Business and enterprise plans
The big providers say they don’t train on business customers’ data by default. Read your plan’s terms to confirm it.
A business plan doesn’t make you compliant
Under a business plan’s data processing agreement, the provider handles data on your instructions, but your firm stays responsible for what goes in. Putting a TFN into a business plan is still your call, and your risk.
Feedback buttons can share a chat
On some tools, a thumbs up or down sends that conversation to the provider for review, even with training off. Give feedback in the chat instead.
Training off doesn’t mean deleted
Providers usually keep chats for a period for safety, abuse checks and legal reasons, even with training off.

Plans, names and settings change often. Check yours before you rely on this.

For beginners

Set your firm up in five steps

  1. Step 1: Choose one tool, on a business plan

    Pick one AI tool for the firm and pay for business accounts. Free accounts are for no client work at all.

  2. Step 2: Switch training off everywhere

    On every account, including the personal ones staff already use. Check chat history and memory settings while you’re there.

  3. Step 3: Write the rule down

    Put the traffic lights above into your AI policy, and walk the team through it. Our free AI policy template is a starting point.

  4. Step 4: Strip client details before you paste

    Swap names and numbers for placeholders first. Our redaction tool does it on your own computer.

  5. Step 5: Tell clients and keep a record

    Say how you use AI in your engagement letter, and keep a list of the tools you use and what they’re approved for.

The rules you already follow

AI doesn’t change your obligations

An AI provider is a third party. The duties you already have still apply when you use one.

  • ConfidentialityThe Code of Professional Conduct in the Tax Agent Services Act says not to disclose a client’s information to a third party without their permission, unless you have a legal duty to.
  • PrivacyWhere the Privacy Act applies to your firm, the Australian Privacy Principles cover how you use and disclose personal information (APP 6), sending it overseas (APP 8) and keeping it secure (APP 11).
  • The regulator’s viewThe OAIC’s guidance on commercially available AI products recommends not entering personal information, particularly sensitive information, into publicly available AI tools.

This is a plain-English summary, not legal advice. Check your own obligations with your professional body or adviser.

Before you approve a tool

Questions to ask any AI provider

Training is only one question. Most providers answer the rest in a trust or privacy centre and a data processing agreement. If they can’t answer, don’t approve the tool.

  1. Who can access our data, including your own staff?
  2. Where is it stored, and where is it processed? The two can differ.
  3. How long is it kept, and can we delete it?
  4. Is it used to train models on our plan, and can we switch that off for everyone?
  5. Which other companies handle it for you, and where are they?
  6. What admin controls and activity logs do we get?
  7. Which security certifications do you hold, such as ISO 27001 or SOC 2?

Keep the answers with your AI policy, so you can show how you stayed accountable for the data.

If client data has already gone in

  1. Don’t panic, and don’t hide it. Tell whoever handles privacy in your firm.
  2. Delete the chat, and note what was shared, when and in which tool.
  3. Check that tool’s terms for how long it keeps data and whether it trains on it.
  4. Treat it like any other possible privacy incident: follow your data breach response plan and assess whether it’s likely to cause serious harm.
  5. Fix the cause: training settings, the policy, or a quick refresher for the team.

Use these next

Please read: use these at your own risk

  • These prompts, guides and tools are general information, not professional, legal, privacy or IT advice. The Firm takes no responsibility for how they are built, changed or used, or for anything that results.
  • AI writes a slightly different tool each time. Whatever you build or download is yours to check, test and approve before you rely on it or use it with client data.
  • Check your AI provider’s current settings and terms for how your data is stored and whether it is used to train its models. Plans and settings change.
  • Bring your own judgement, and follow your firm’s AI policy and your professional and privacy obligations.

Questions firms ask

Is it safe to put client data into ChatGPT?

Don’t put identifiable client data into any general AI chat. On a business plan with training off, the risk is lower, but the data still leaves your firm and goes to a third party, usually overseas. Remove client details first, and follow your firm’s AI policy.

Does turning off training make an AI tool safe for client data?

It helps, but it isn’t enough on its own. Providers usually still keep chats for a period, and your confidentiality and privacy duties still apply to sending the data at all. Strip client details first.

I’m a sole practitioner. Do I need a team or business plan?

Not necessarily. A personal paid plan with training switched off can be enough if you keep identifiable client details out and redact first. A business plan adds a data processing agreement, admin controls and training off by default, but some have a minimum number of seats, so compare the cost with how you’ll actually use it. Either way, your firm stays responsible for what goes in.

Can staff use their personal ChatGPT or Claude accounts for work?

Not for anything to do with clients. Personal plans often train on chats unless the setting is switched off, and the firm can’t see or control those accounts. Give staff an approved business account instead.

What should we do if someone pasted client data into an AI tool?

Tell whoever handles privacy, delete the chat, and record what was shared, when and where. Check the tool’s retention and training terms, then follow your data breach response plan and assess whether it’s likely to cause serious harm.

Staff have been trained on our AI policy but still do the wrong thing. What now?

Make the safe way the easy way. Give everyone an approved business account, ask your IT provider whether free AI sites and personal accounts can be blocked on firm devices, use the tool’s admin controls and activity logs, run short refreshers, and make it safe to own up to a mistake quickly.

I don’t understand how AI works behind the scenes. How do I avoid a breach by accident?

You don’t need to know the technology to stay safe. Follow one rule: client details stay out of the chat unless your firm has approved the tool for them. Remove names and numbers first, and ask before trying anything new.

Does our insurance cover using AI?

Ask your broker, and get the answer in writing. Tell them which AI tools you use and how, and ask whether your professional indemnity cover extends to work prepared with AI and whether your cyber cover responds to an incident involving an AI provider. Policies differ.

Do we need to tell clients we use AI?

It’s good practice, and it builds trust. Say in your engagement letter how you use AI, that a person reviews the work, and that you don’t put identifiable client data into public AI tools. The redaction tool page has wording you can adapt.