Retirement projections and AI data leakage: what a cash flow spreadsheet gives away in ChatGPT or Claude.

A retirement projection is the spreadsheet an adviser builds to show a couple whether they can stop work: ages, balances, contributions, spending, and the Age Pension they might receive later. Advisers paste it into ChatGPT or Claude to check the formulas, summarise the scenarios or find the row that does not add up. It holds more than the grid shows. It names the clients, carries their balances to the dollar, records assumptions about a redundancy and a health condition, and links to other tabs. This page marks the items in an invented projection that the Code of Ethics and the Privacy Act bear on, and shows the same model checked on Ermos inside the practice.

The retirement projection spreadsheet

Built before an annual review or a retirement strategy meeting. Standard 6 of the Code of Ethics requires an adviser to take into account the broad effects of the client acting on the advice and to actively consider the client’s broader, long-term interests and likely circumstances, which is what a projection models.

An invented projection. All figures are illustrative, including the Age Pension estimates. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the practice

1. 1
   The clients, in the tab names

   Two named individuals in a workbook title, with ages in the columns. Personal information under the Privacy Act, where the Act applies to the practice. Privacy Act guide.

2. 2
   Opening super balances to the dollar

   The base of the whole projection: two people’s superannuation as at the review. Financial information the practice keeps in client records under Standard 8, and the starting point from which every later row can be rebuilt. Privacy Act guide.

3. 3
   A redundancy and an income assumption

   That a named client will lose his job in December, the expected package and the salary he is assumed to earn next. Confidential employment information, given to the adviser in confidence. Privacy Act guide.

4. 4
   Goals and the retirement date

   When the clients want the mortgage cleared, when they plan to stop work and what they want to spend. The inputs Standard 6 asks the adviser to consider when weighing the clients’ long-term interests.

5. 5
   A health scenario named in a tab

   A scenario in which one client stops work at 55 because of her thyroid condition. The tab name alone reveals a health concern, and a scenario built on it. Health information is sensitive information under the Privacy Act. Patient data guide.

6. 6
   A parent’s estate and care costs

   A reference to a named relative, her possible inheritance and uncertain care costs. Personal information about a person who is not the client, and a family matter. Privacy Act guide.

7. 7
   The Age Pension estimates

   Services Australia says the Age Pension age is 67 for people born on or after 1 January 1957, which is why the rows begin in 2046 and 2048. The dollar figures depend on the income and assets tests and rates in force in future years. The estimates here are invented and a person must check any such figure against Services Australia before it goes into advice.

8. 8
   The model’s assumptions

   Net return, contribution rates and who contributes what. The practice’s modelling choices, written for one household, with the clients’ contribution amounts in them.

9. 9
   A formula that points at another tab

   A formula built from named cells in an assumptions tab. Formulas can also point at other workbooks. This one holds no client data of its own.

10. 10
    File properties and what the sheet does not show

    The adviser’s name and the model version. Spreadsheet files can also carry hidden rows and columns, cell comments, tracked changes and properties such as the author and the practice, none of which show in the grid but all of which travel with the file.

What happens in Frontier Models (ChatGPT, Claude)

• The whole document is transmitted to the provider’s servers. On personal plans the user cannot choose where it is stored or processed, and for the major providers that has generally meant outside Australia, under that country’s legal process. Some business plans offer storage in Australia, but the provider still receives the content.
• On consumer tiers, prompts can be retained and used to train future models.
• The provider’s staff and systems can reach the content under the provider’s own policies.
• Once sent, the practice no longer controls it. Retention and deletion happen on the provider’s terms.

The same document on Ermos

Part of AI data leakage in financial advice practices. See how this works in practice: Review preparation on Ermos.

Common questions