AI data leakage, one document at a time.

Most advice about AI security stays general. This series does the opposite. For each profession we take the documents it produces every week, mark every item it is obliged to protect, and show exactly what leaves when the document goes into ChatGPT or Claude, which obligation each item touches, and how the same document behaves on Ermos Edge and Ermos Dominion.

By profession

Accountants

18 documents marked up, with the obligations each item touches.

See what leaves → Law firms

18 documents marked up, with the obligations each item touches.

See what leaves → Financial advisers

17 documents marked up, with the obligations each item touches.

See what leaves → Insurance brokers

17 documents marked up, with the obligations each item touches.

See what leaves → Aged care

17 documents marked up, with the obligations each item touches.

See what leaves → Mortgage brokers

10 documents marked up, with the obligations each item touches.

See what leaves → Healthcare

10 documents marked up, with the obligations each item touches.

See what leaves → General practitioners (GPs)

10 documents marked up, with the obligations each item touches.

See what leaves → Dental practices

10 documents marked up, with the obligations each item touches.

See what leaves → Conveyancers

10 documents marked up, with the obligations each item touches.

See what leaves → Migration agents

10 documents marked up, with the obligations each item touches.

See what leaves → Real estate agents

10 documents marked up, with the obligations each item touches.

See what leaves → NDIS providers

10 documents marked up, with the obligations each item touches.

See what leaves → Recruiters

10 documents marked up, with the obligations each item touches.

See what leaves →

By document type

Letters

28 documents across 14 professions, marked up item by item.

See them → Agreements

9 documents across 8 professions, marked up item by item.

See them → Emails

31 documents across 14 professions, marked up item by item.

See them → Notes

13 documents across 10 professions, marked up item by item.

See them → Reports and memos

20 documents across 12 professions, marked up item by item.

See them → Forms

19 documents across 14 professions, marked up item by item.

See them → Spreadsheets

25 documents across 14 professions, marked up item by item.

See them → Records

14 documents across 11 professions, marked up item by item.

See them → Transcripts

9 documents across 9 professions, marked up item by item.

See them → Text and chat messages

7 documents across 7 professions, marked up item by item.

See them →

What the pages show

Each page takes one real document type, invented in every detail, and marks every item the business is obliged to protect. Hover or tap an item for the obligation it touches and a three-way comparison: what happens to it in Frontier Models (ChatGPT, Claude), on Ermos Edge, and on Ermos Dominion. The full analysis sits underneath for reading in order.

Common questions

What happens to a document when it is pasted into ChatGPT or Claude?
The whole text travels to the AI company's servers, is processed there to produce a reply and is kept in the user's conversation history. As at October 2026, whether a conversation on a personal plan (ChatGPT Free, Plus or Pro; Claude Free, Pro or Max) can be used to train the models depends on a setting each user controls. Business plans (ChatGPT Business and Enterprise, Claude Team and Enterprise) and the developer APIs sit under commercial terms, and OpenAI and Anthropic each say they do not train on that content by default (OpenAI, Anthropic). On either product, the thumbs up and thumbs down buttons can send a whole conversation back for training. Every item marked on these pages is part of what travels.
Do ChatGPT and Claude keep the documents people paste in?
Deleting a chat removes it from the user's history, but OpenAI and Anthropic each say they can keep a deleted chat on their systems for up to 30 days, and longer where safety or legal obligations apply (OpenAI, Anthropic). Both products also have memory features that can carry details from one conversation into later ones, and on a personal account each user decides whether they are switched on (ChatGPT, Claude). Both companies can also review content that is flagged for safety or policy reasons. For a business, the practical point is that a deleted chat does not prove a client's document has gone from the AI company's systems.
What does the OAIC say about putting client information into ChatGPT or Claude?
The OAIC is the federal privacy regulator. The Office of the Australian Information Commissioner (OAIC) recommends in its guidance on commercially available AI products that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools, because of the significant and complex privacy risks. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. Removing names rarely makes a document anonymous. Information stays personal information while the person is reasonably identifiable, and the OAIC warns that information can be at risk of re-identification even when it has been de-identified or anonymised. The guidance is a regulator's view rather than law, and it shows how the Australian Privacy Principles are being read when staff use these tools.
Do business plans such as ChatGPT Business and Claude Team fix the problem?
On a personal plan the user cannot choose where conversations are stored or processed. Some business plans let an organisation choose to store data at rest in Australia (OpenAI), which still leaves the content with the AI company. A business plan changes the default on training, but it does not change who receives the content. The question for the business becomes whether sending a client's document to that AI company is something the client agreed to or would reasonably expect. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product.
How does Ermos avoid sending documents to ChatGPT or Claude?
Ermos is built so the document stays under your control. On the Dominion AI unit it is drafted inside your office and none of it is sent to an AI provider. On Ermos Edge your documents and index stay on your own instance in an Australian data centre, and only the question and the excerpts the user is permitted to see go to a named Australian provider that is contractually barred from keeping them or training any model on them. Every document page in this series shows the same document item by item on both products.
Are the example documents real?
No. Every document is invented, including the names, figures and references. Identifiers use published example values, such as the ATO's example tax file number, so nothing can match a real person or entity.
Why name ChatGPT and Claude?
Because they are the tools staff reach for. Both are capable products from reputable providers, and our comparison pages say so. These pages are about what leaves your business when a document goes into any frontier model, whoever runs it.

See Dominion answering from your own documents.

A demonstration takes 30 minutes, uses no client data, and comes with no obligation.

Book a demo
Keep reading
Compliance guides What is Resident AI The Dominion AI unit