Serviceability spreadsheets and AI data leakage: what a borrowing calculator gives away in ChatGPT or Claude.

A serviceability spreadsheet shows how close a household is to its limit. It lists net income, overtime shaded by the lender’s policy, living expenses against a benchmark, every card limit and the repayment at a buffered rate, and it ends on the surplus. Brokers paste it or upload the file to ChatGPT or Claude to find a formula error, to test scenarios and to explain the result to the client. This page marks every cell in a routine serviceability sheet that the National Consumer Credit Protection Act and the Privacy Act bear on, and shows how the same sheet is checked on Ermos inside the brokerage.

The serviceability spreadsheet

Built after the documents are in and before a lender is chosen, and rerun for each lender’s policy. It supports the preliminary assessment that section 115 requires within 90 days before the credit assistance, and the assessment of whether the contract would be unsuitable under section 118.

fx=D5-SUM(D6:D10)
ABCDE
1ItemDetailDeclared / monthAssessed / monthCheck
2Net salary, 9,2829,282Payslips 25 Sep, 9 Oct
3Overtime, Daniel488390YTD on payslip
4Net salary, Riverbend Allied Health, 0.8 FTE, $84,0005,5285,528Payslips 24 Sep, 8 Oct
5Total net income15,29815,200=SUM
6Living expenses5,1505,620Statements: $5,350
7ChildcareNoah, three days a week, gap fee1,0901,090Centre statement
8Car loan612612Statement 8 Oct
9Credit cards1405703.8% of limits
10New loan repayment5,1436,916Declared at actual rate
11Surplus per month3,163392Passes, narrowly
12Debt to income3.68Policy limit 6.00
13Lender policySouthgate policy v3
14MethodStandard
ServiceabilityLender policyScenarios

An invented serviceability spreadsheet. The lenders and their policy parameters are invented. Hover or tap any highlighted cell for the analysis; the full list is below.

What leaves the brokerage

  1. 1
    The applicants

    Named in the row labels, the file name and usually the sheet tab. The key that ties every figure to two real people. Privacy Act guide.

  2. 2
    Employer and base salary

    A named employer and a salary. Financial information collected because section 117 requires reasonable steps to verify the consumer’s financial situation. Privacy Act guide.

  3. 3
    Overtime and its shading

    How much the household earns beyond base pay and how a lender’s policy treats it. The figure that often decides whether an application passes. Privacy Act guide.

  4. 4
    Declared expenses against the benchmark

    What the clients say they spend, what a benchmark says a household like theirs spends, and the figure checked against statements. The brokerage’s inquiry into the consumer’s financial situation, laid bare. Privacy Act guide.

  5. 5
    Existing debts and card limits

    A named credit union, balances and the limits on two cards. The commitments a lender counts in assessing whether the contract would be unsuitable under section 118. Privacy Act guide.

  6. 6
    The new loan and the assessment rate

    The proposed loan, its term and the buffered rate the repayment is tested at. The deal and the stress test in one cell. Privacy Act guide.

  7. 7
    The surplus and the scenario

    A household that passes by a few hundred dollars, and the broker’s plan to widen the margin. Highly personal to the client, and a number the broker must check by hand, since a model can add up wrongly. Privacy Act guide.

  8. 8
    Debt to income

    A ratio built from the loan, existing debts and gross income. Reveals gross income and total debt in one figure. Privacy Act guide.

  9. 9
    Lender policy parameters

    A buffer, an overtime rule and a DTI limit. Generic and free of client data, though some lender policy documents sit behind a broker portal’s terms of use.

  10. 10
    The method

    A standard repayment formula and the surplus arithmetic. No client figures. Spreadsheet files can also carry hidden sheets and columns, cell comments, formulas that pull from other sheets and file properties, all of which travel with an uploaded 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 brokerage no longer controls it. Retention and deletion happen on the provider’s terms.

The same document on Ermos

Ermos Dominion

The document is drafted on a unit in your office, from your own templates and the client file. Every marked item stays on the unit, which keeps working with the internet switched off. Nothing in it is transmitted to an AI provider, so there is no disclosure to analyse: the confidentiality work is the same as for your file server.

Ermos Edge

Your documents, the index and your access boundaries stay on your own instance in an Australian data centre. When a staff member asks for this draft, only their question and the excerpts they are permitted to see go to a named Australian provider, who is contractually barred from keeping the content or training any model on it. Nothing is processed outside Australia.

Frontier Models (ChatGPT, Claude)Ermos EdgeErmos Dominion
Where the document is processedProvider servers, generally offshore on personal plansYour instance in an Australian data centre; answers generated in AustraliaThe unit in your office
Who can read itProvider staff and systems, under policyYour staff, within the boundaries you set. The named Australian provider processes only permitted excerpts, under contract, and keeps nothingYour staff only
Is it keptOn the provider’s termsKept on your instance; the provider is barred from retaining itStays on the unit
Can it train a modelOn consumer tiers, yesNo, by contractOnly your own unit learns from it, on your hardware

Part of AI data leakage in mortgage brokerages. See how this works in practice: Lender policy lookups on Ermos.

Common questions

Do ChatGPT and Claude use the information in a serviceability spreadsheet?
Whatever is pasted or uploaded from the sheet goes to the AI company's servers and is processed there to produce a reply. 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. The sheet puts two named applicants beside their net income, the overtime shading, every card limit and a surplus of $392 a month.
Is it safe to paste a serviceability spreadsheet into ChatGPT or Claude if we delete the client names?
No, because the figures describe one household even without the names. 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. An $868,000 loan tested at 8.89 percent, a named credit union's car loan and two card limits narrow the field to one couple. 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.
Can ChatGPT or Claude be trusted with the calculations in a serviceability spreadsheet?
Not without a check. Language models can produce fluent text with wrong figures, wrong totals and formulas that look right but are not. A qualified person should check every item against the file. In this sheet the overtime shading, the buffered repayment, the expenses against the benchmark and the $392 surplus would do the most harm if wrong, because they decide whether the contract is unsuitable under section 118.
How can a mortgage brokerage use AI for a serviceability spreadsheet without sending it 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. The sheet is produced from the client file and the brokerage's own records. The broker checks every figure and the surplus by hand before relying on it.
Can AI do the serviceability calculation?
It can add up, but the obligation stays with the licensee. Sections 115 to 118 of the National Consumer Credit Protection Act require the licensee to make the inquiries, verify the consumer's financial situation, make the preliminary assessment and assess unsuitability. A figure produced by a tool is only as good as the inputs and the check the broker applies.
What else travels with a spreadsheet file?
More than the visible cells. Spreadsheet files can carry hidden sheets and columns, cell comments, formulas that refer to other sheets, tracked changes and file properties such as the author. Uploading the file sends all of it. Pasting only the visible cells sends less, but the visible cells are the sensitive part here.
How long is a preliminary assessment good for?
Section 115 requires the preliminary assessment and the inquiries to have been made within 90 days before the day the licensee gives the credit assistance, and the assessment must specify the period it covers. After that the broker needs to redo them. The spreadsheet is the working paper behind that assessment.

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
AI data leakage in mortgage brokerages AI for mortgage brokers The Privacy Act, the APPs and AI All professions