You upgraded to ChatGPT Business or Claude Team. Who holds your client's data now?
A business plan fixes the training question. It does not answer where client data sits, how long it is kept, or which country's law applies. Here is the checklist, and how a dedicated, Australian-hosted AI employee handles each point.
Short Answer
Upgrading to ChatGPT Business or Claude Team stops your client data being used to train the vendor's models by default. It does not decide where the data is stored, how long it is kept, what happens to flagged chats, or which country's law can compel disclosure. Those four questions are the ones to answer before client data goes in.
Last reviewed: October 2026
Key takeaways
- Business tiers solve the training question. Anthropic's commercial terms say it "may not train models on Customer Content from Services."
- Retention is a separate question. Deleted Claude conversations leave back-end storage within 30 days, and chats flagged by automated systems can be kept for up to 2 years.
- A US vendor is subject to US legal process. Anthropic's commercial terms allow disclosure "to the extent it is required by law, or court or administrative order."
- Agentive stores client data at rest in Australia and runs inference on Amazon Bedrock under zero data retention, in Australia by default.
- Inference can be routed to Amazon Bedrock in the United States in two stated cases, unless the customer has made the Australia-only election.
Plenty of bookkeepers and BAS agents have taken the sensible step of moving off a free chatbot and onto a paid business plan. It is the right first move. It is also where many practices stop asking questions. This post sets out what a business plan settles, what it leaves open, and how a dedicated, Australian-hosted AI employee handles each open point. It reads the vendors' own published terms rather than summaries of them.
What does a business AI plan actually fix?
It fixes training. Anthropic's privacy centre states: "By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models." Its commercial terms add that the customer "retains all rights to its Inputs" and "owns its Outputs."
There is one exception to note. The same page says that if you report feedback, for example with the thumbs up or down button, or otherwise choose to allow use of your data, "then we may use your chats and coding sessions to train our models." A practice that lets staff click feedback on a chat containing a client's figures has opted that chat in.
OpenAI publishes comparable commitments for its business products on its enterprise privacy page. We have not quoted OpenAI's wording here because its pages were not readable by our tools when this post was written, so read that page and the data processing addendum yourself before relying on any summary, including ours.
What stays open after you upgrade?
1. How long is client data kept?
Anthropic's retention page says a deleted conversation is "deleted from our back-end storage systems within 30 days." It also says: "We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years if your chat is flagged by our automated trust and safety systems as violating our Usage Policy." A client file can be flagged for reasons that have nothing to do with the client, and you would have little visibility of it.
2. Which country's law applies?
Anthropic's commercial terms let a recipient disclose confidential information "to the extent it is required by law, or court or administrative order", and say it will, "except where expressly prohibited, notify Discloser of the required disclosure promptly." That is a reasonable clause. It also means a US vendor can be compelled under US process, which is a point worth disclosing to clients who expect their data to stay under Australian law.
3. Where is it stored and processed?
The vendor pages we could read do not state a data location for Team plans, and a business chat plan is a shared product, not a server of your own. If your engagement terms promise Australian handling, check that the vendor will put the location in writing before you rely on it.
4. What can the tool reach?
Connectors to Xero or MYOB on chat plans are typically read-only, so the assistant can look but not post, reconcile or file. That limits risk and also limits usefulness. Check the current connector documentation for the exact scope on your plan.
How does a business chat plan compare with a dedicated AI employee?
| Question | Business chat plan | Agentive AI employee |
|---|---|---|
| Training on your data | Off by default (Anthropic). Feedback clicks can opt a chat in. | None. Customer data is not used to train or improve any model. |
| Where data at rest sits | Vendor's shared platform. Check the written terms. | Your own single-tenant server in AWS Sydney or Melbourne. |
| Retention | 30 days after deletion. Up to 2 years if flagged (Anthropic). | Zero data retention on Amazon Bedrock. Conversations live only on your instance and its Australian backups. |
| Governing law for disclosure | US vendor, so US legal process. | Australian company, with AWS as the infrastructure provider. |
| Where inference runs | Vendor's choice. | Australia by default. The United States only if the model is unavailable locally or capacity is short, with notice. Never, if you elect Australia-only. |
| What it can do in Xero or MYOB | Often read-only connectors. | Works inside your systems with approval gates and an audit log of every action. |
How does Agentive handle each point?
Each Agentive AI employee runs on a dedicated single-tenant server, so there is no shared memory pool or shared queue with another customer. Data at rest, which covers the instance, memory, conversations, files and backups, is stored in Australia on AWS. Inference runs on Amazon Bedrock under a zero data retention arrangement with AWS, so no prompt or response is written to storage by AWS or shared with a model developer. Your data is sent only to Amazon Bedrock and never to a model developer.
We should be precise about one limit, because precision is the point of this post. Inference runs in Australia by default. Under our Terms, it can be routed to Amazon Bedrock in the United States where the model we have chosen is not available in Australia, or where Australian capacity is short. Your data at rest stays in Australia, the same zero data retention and no-training terms apply, and we notify you. If you need inference that never leaves Australia, you can make the Australia-only election, and we will not route you elsewhere.
What should a practice do this week?
- Read the current terms and data processing addendum for every AI tool staff use, and note the retention and law-of-disclosure clauses.
- Switch off feedback buttons and any data-sharing options on business plans.
- Ask each vendor, in writing, where your data is stored and processed.
- Check your engagement terms and privacy statements describe your AI use accurately. Our TPB guidance post covers what to disclose.
- Decide which client data goes into a general chat plan at all, and which belongs in a dedicated environment. Our post on putting client data into AI is a good starting point.
Bring One Client Agreement and We Will Map the Data Flow
Agentive builds dedicated AI for Australian bookkeepers, BAS agents and small businesses. Each AI Employee runs on a dedicated single-tenant server in Australia, client data at rest is stored onshore and never used to train models. Bring one client agreement to the call and we will walk through where each piece of data goes.