Trial balances and AI confidentiality: what a year-end spreadsheet gives away in ChatGPT or Claude.

A trial balance looks like arithmetic, which is why it is so easy to paste into ChatGPT or Claude: check that it balances, suggest the adjusting journals, explain the movement to the client. It is also the whole business in one grid. The account names carry family loans, related-party rent and the contractors the practice pays, and the reviewer’s notes beside them say what the firm suspects. This page marks every item in a year-end trial balance that AI confidentiality rules bear on, shows which parts are only generic headings, and shows how the same review runs on Ermos inside the practice.

The year-end trial balance

Extracted from the ledger once the year-end bookkeeping is complete, then annotated by the preparer and the reviewing manager before the accounts are drafted.

Trial balance

fx=SUM(C2:C13)
ABCDE
1Debit ($)Credit ($)Reviewer note
2100Fees income2,486,300
3210Contractor dentists612,400
4220Wages and superannuation820,150
5230Supplies and other expenses603,200
6240Rent96,000
7510Cash at bank260,700
8540Plant and equipment428,600
9560Loan - Marco Marchetti142,000
10620Loan - Giuseppe Marchetti200,000
11640Westpac term loan120,000
12660GST and PAYG payable88,750
13900Opening equity68,000
14Total2,963,0502,963,050
TB 30-Jun-26JournalsNotes

An invented trial balance. The account ending digits are placeholders. Hover or tap any highlighted cell for the analysis; the full list is below.

What leaves the firm

  1. 1
    The entity and the year end

    The company, its trading name and the balance date. Names the business behind every number in the grid. A workbook also carries author and last-saved-by fields and any hidden tabs or columns when it is shared as a file. APES 110 confidentiality guide.

  2. 2
    Practice fees for the year

    Total revenue and the trend. Confidential financial information acquired in the professional relationship, covered by APES 110 R114.2. APES 110 confidentiality guide.

  3. 3
    Named contractors and what they were paid

    Two named dentists and what the practice paid for their services. Personal information, and information about how the practice engages its dentists. Privacy Act guide.

  4. 4
    Wages to a family member

    A payment to a relative of a director with the reviewer’s doubt about whether it is at market rate. The doubt is the firm’s own judgement and has not yet been put to the client. Checking AI output guide.

  5. 5
    Related-party rent and the reviewer’s note

    Who the landlord is, that rent was below market for two quarters, and that a top-up journal is pending. A related-party disclosure point for the company, on a third party’s servers before it is resolved. TPB Code guide.

  6. 6
    A loan to a shareholder with no agreement

    A balance, no agreement on file and no repayments. Division 7A of the Income Tax Assessment Act 1936 can treat a loan from a private company to a shareholder or an associate as a dividend, which makes this a live exposure in the client’s name. TPB Code guide.

  7. 7
    A loan from an 81-year-old and a family request

    The lender’s age, the amount, the unpaid interest and a request that one brother not see the line. A confidence inside the family that the firm holds, and personal information about an elderly man. Privacy Act guide.

  8. 8
    Bank facilities and accounts

    The lender, the personal guarantees and the tail of an operating account number. Information about the owners’ own exposure, and a start for anyone building a profile of the business to impersonate it. APES 110 confidentiality guide.

  9. 9
    Unpaid PAYG and a payment plan

    An unpaid tax liability and that the client is seeking terms from the ATO. A compliance position the client has not yet taken to the regulator. TPB Code guide.

  10. 10
    The chart of accounts headings

    Standard account codes and headings such as Code, Account, Debit and Credit. Generic wording found in any ledger, with no client data in it.

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 firm 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 accounting firms. See how this works in practice: Workpaper review on Ermos.

Common questions

Do ChatGPT and Claude use the information in a year-end trial balance?
Yes, all of it goes to the AI company's servers and is processed there to produce a reply, whether it is pasted or uploaded. 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. In this trial balance that includes a loan from 81-year-old Giuseppe Marchetti, a Westpac loan guaranteed personally by Lena and Marco, and $41,200 of unpaid June quarter PAYG.
Is it safe to paste a year-end trial balance into ChatGPT or Claude if we take the client's name off it?
No, because the account names describe one family business even without the heading. 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. Rent paid to a family trust, a loan from an 81-year-old with unpaid interest and two contractor dentists paid $612,400 point to one practice and one household. 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.
Can ChatGPT or Claude be trusted to check a year-end trial balance?
Not without a person checking every line. A language model can produce fluent text with details that are not in the source, wrong figures, wrong dates or references that look real but are not. In a spreadsheet that also means wrong formulas, wrong totals and mis-copied rows. A qualified person should check every item against the ledger and the client file. The costliest errors here would be a total that no longer reaches 2,963,050 on both sides, a mis-copied loan balance or a wrong unpaid PAYG amount.
How can an accounting practice use AI for a year-end trial balance 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. Ermos drafts the review notes from the client ledger and the practice's own trial balance template. A qualified accountant checks every balance and signs it off before the accounts are drafted.
Does a trial balance contain personal information?
Often, yes. Account names such as a loan from a named person, contractor fees by name or wages to a relative are personal information about identifiable individuals. The Privacy Act applies to businesses above its small business turnover threshold and to some smaller ones, so whether it applies to a given practice depends on the facts, but APES 110 confidentiality applies to the information either way.
What is Division 7A in plain terms?
It is a set of rules in the Income Tax Assessment Act 1936 aimed at private companies making tax-free distributions to shareholders or their associates through payments, loans or forgiven debts. A loan with no complying agreement can be treated as a dividend. A trial balance showing a shareholder loan is therefore showing a potential tax exposure as well as a balance.
Can AI check that a trial balance balances?
It can add columns, but it can also misread a figure or invent a total, so a person has to check the result. TPB(GS) 55/2026 says tax practitioners should verify and review AI generated content for accuracy and that each step should be documented. The risk of an arithmetic mistake stays the same wherever the work is done.

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 accounting firms AI for accountants APES 110 and AI All professions