Is my data safe to use with this AI tool?
A decision guide for confidential information.
In brief
- AI adds three questions to your usual due diligence on any app: Does it train on my data? Where does it store my data? Who does it share my data with?
- Your tools fall into three buckets: the AI in your office suite, the AI built into your business apps, and standalone tools you bring in.
- Your data policy and professional obligations, including client confidentiality, come first.
By Inbal Rodnay
Is my data safe to use with this AI tool?
Part of this you already know how to handle. You have trusted cloud software with your data for years: files on OneDrive, Google Drive or Dropbox, records in your accounting software, and emails and attachments in Outlook or Gmail. Handing data to another company to store is not new, and you already have ways to judge it.
What is new is the risk that the AI learns from it: not just holding your data, but training on it, and potentially drawing on what it learned to answer someone else’s question.
This article untangles it: the three questions to ask about any tool, and the three buckets your tools fall into.

Three questions to ask
Whatever the tool, there are three questions you need to ask, on top of the usual ‘cloud app’ questions.
Get the answers from the vendor and keep them in writing. Some vendors publish them on their websites; for others, you will need to ask. Don’t hold back. Vendors field these questions every day.
Here are the questions:
- Does it train on my data? On a work-supplied business account, the major providers say no. But read the small print. For example, on some business plans, rating a chat with a thumbs up or down sends the whole conversation to the vendor. Anthropic, one major provider, keeps rated conversations for up to five years and may use them for training. Ask your administrator or IT provider whether this feature can be switched off.
- Where does it store my data? Ask where it is stored, where it is processed and how long it is kept. Some vendors commit to keeping data in Australia or New Zealand; others process it offshore or keep it for a period after you delete it.
- Who does it share my data with? AI vendors and apps with built-in AI often use other companies to process data on their behalf. These companies are called sub-processors. An app that adds AI capabilities typically passes your data to an AI provider behind the scenes, a third party you never chose. You want to know who it is and what it is allowed to do with your data.
Confirm the answers for the specific tool and business account you use.
Where your tools sit
The AI in your office suite. Microsoft Copilot inside Microsoft 365, Google Gemini inside Google Workspace. You are not starting from scratch. The day you put your documents, emails and files onto Microsoft or Google, you decided to trust them with your data. The AI in that suite usually rides on the same trust and protections, with some exceptions covered below. This guide uses Microsoft as the example from here; the same questions apply to Google.
On a business account, Microsoft makes two key commitments.
The first is protection: with enterprise data protection, your Copilot prompts and responses are covered by the same contractual commitments as the rest of your Microsoft 365 business data.
The second is training: your prompts, responses and the work data Copilot accesses are not used to train foundation models. Web search and third-party agents can have different data-handling arrangements, so check the detail.
To verify these commitments, look for the green shield at the top of Microsoft 365 Copilot Chat, or read Microsoft’s guide, Data protection when using Microsoft Copilot Chat for work or school.
Claude inside Copilot is a special case. Copilot draws on more than one AI model to answer your prompts, including models from OpenAI and Anthropic (Claude). Anthropic provides most of its Claude models to Microsoft as a sub-processor, under Microsoft’s contract and data protection terms. The difference is where your data goes. Microsoft says Claude models are currently excluded from its in-country processing commitments where those apply, so your data may be processed outside Australia or New Zealand. Microsoft has switched Claude models on by default for most commercial customers outside the EU, the UK and the European Free Trade Association (EFTA) countries. If offshore processing is a concern, ask your Microsoft 365 administrator or IT provider to review the setting. One exception: models labelled “Anthropic models with Data Retention” fall under Anthropic’s own terms, not Microsoft’s, and stay off until an administrator turns them on.
The AI built into your business apps. Your accounting, ERP, practice management and tax software are all adding AI features. That does not mean the vendor has built its own AI. Usually, the app is sending your prompt or your data to a model provider behind the scenes, such as OpenAI, Anthropic or another AI service. So the due diligence is not only about the app you know. It is also about the AI provider sitting underneath it.
Ask the vendor the AI questions directly. Ask which AI model or provider they use as a sub-processor, and then it’s the same three questions: does it train on my data, where does it store my data, and who does it share my data with?
Standalone tools you bring in. A standalone ChatGPT, Claude or Perplexity is a different case again. These tools need their own assessment. If your organisation has not already approved the tool for the information and task, assess it as you would any new third-party provider.
Watch for the paid trap. Paying for a personal plan does not make it a business plan. On a personal plan, whether your chats are used to train the model is usually set by the individual user, not the organisation, and on some plans it is switched on by default. Personal plans also do not come with the business terms your organisation needs.
So do your due diligence. Read the terms, check that a business plan or agreement commits to no training and controlled storage, then decide. The same applies to any new practice, advisory or service-line tool. For help choosing, see: Five questions to choose your primary AI tool.
No-go situations, whatever the tool
- Client or confidential information in a personal AI account, whether free or paid.
- Information your policy classifies as restricted, unless the tool is approved for that level.
- Client information shared without the permission your obligations require. Australian registered tax agents, see TPB(GS) 55/2026.
- Tax file numbers, IRD numbers or other government identifiers, without checking your obligations first. In Australia, the Privacy (Tax File Number) Rule 2015 may apply.
- Preview or beta features with client or confidential information, until you have checked their terms.
Let your policy and your obligations lead
Where your organisation has a data classification policy, it takes precedence. Classify the information under your own policy first, then use this guide to decide where it can go.
A tool’s terms are only half the question. Your professional obligations still apply: the confidentiality principle in the Code of Ethics that applies to you, privacy law (including the rules on disclosing personal information overseas: APP 8 in Australia, IPP 12 in New Zealand) and, for registered tax agents in Australia, the Tax Practitioners Board’s Code of Professional Conduct.
The TPB’s guidance, TPB(GS) 55/2026, notes that entering client information into an AI tool may be a disclosure to a third party that needs your client’s permission, depending on how the tool is set up. If you are in public practice, check that your engagement letters cover it. If you are in-house, check with your privacy officer or legal team how this applies to the information you handle. See: What your AI policy has to cover.
Get it right, then trust your team
Asking professionals to anonymise every piece of work before using an approved AI tool is rarely workable in practice, and it can undermine the productivity the tool is meant to deliver. The answer is not fear. It is one clear, well-made decision about which tools are approved for which information and tasks, communicated to your team, so everyone can get on with the work. Revisit that decision whenever the tool, its terms or the way you use it changes. The goal is a trusted, secure AI tool for every team member who needs one.
Sources
Microsoft, Data protection when using Microsoft Copilot Chat for work or school
Microsoft, Privacy and protections
Microsoft, Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat
Microsoft, Anthropic models in Microsoft Online Services
Anthropic, Is my data used for model training?
Disclaimer
This guide provides general guidance only and is current as at October 2026. Tool names, features and capabilities were correct at that date but may change. The publication therefore focuses on general principles rather than specific feature lists.
This guide is not intended to be comprehensive and is not a substitute for professional, legal or other advice tailored to your circumstances. Use it alongside your organisation’s artificial intelligence policies and your applicable legal, regulatory, professional and ethical obligations. Check current vendor terms, product information and your own obligations before acting on or relying on the information.
While Chartered Accountants Australia and New Zealand (CA ANZ) seeks to ensure that the information is accurate and current, it does not guarantee its accuracy, completeness, currency or suitability for any particular purpose. To the fullest extent permitted by law, CA ANZ disclaims all warranties and liability for any direct, indirect, special or consequential loss or damage arising out of or in connection with any inaccurate statement, opinion, advice or information contained in this publication, any omission from it, or the use of or reliance on this [guide/document] or any material it contains.
© 2026 Chartered Accountants Australia and New Zealand ABN 50 084 642 571