Claims spreadsheets and AI data leakage: what an NDIS invoice sheet gives away in ChatGPT or Claude.

Every NDIS service keeps a spreadsheet that turns shifts into claims. Staff paste it into ChatGPT or Claude, or upload the file, to find the rows that were rejected, check a formula or draft the email to a plan manager. The sheet is a table of identifiers: each participant, the NDIS number the NDIA uses to match claims, the support delivered, the hours and a status column that reveals which shifts are being questioned and why. This page marks every item in a month’s claims sheet that the Practice Standards and the Privacy Act bear on, and shows what travels with the file, then how the same checks run on Ermos inside the service.

The claims and invoice spreadsheet

Maintained by the service administrator through the month and used to submit claims. NDIA guidance on invoicing and record keeping expects providers to keep records of the supports they deliver, such as case notes, rosters and service agreements, to support the claims they make.

fx
ABCDEFGH
1ParticipantNDIS no.Support itemDateHoursRateTotalClaim status
219 Oct2.0$136.80Paid
320 Oct1.0$68.40
426 Oct4.0$273.60Submitted
519 Oct6.0$410.40Submitted
621 Oct6.0$410.40Paid
716 Oct6.0$410.40
816 Oct3.0$205.20Paid
923 Oct3.0$205.20
1024 Oct4.0$384.80Submitted
1127 Oct4.0$273.60Submitted
Oct-26 claimsRatesNotes

An invented claims sheet. The NDIS numbers are placeholders, and the rates are invented and are not NDIS price limits. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the service

  1. 1
    Participant names

    Every participant in the month, one row per shift. Personal information the Information Management standard expects to be kept confidential, repeated down a column. Privacy Act guide.

  2. 2
    NDIS numbers

    The number the NDIA uses to match a payment request to a participant’s plan. A column of them beside names makes impersonation and claims fraud easier. Treat it like a column of bank account numbers. Privacy Act guide.

  3. 3
    The support each person receives

    Self-care, community access, daily living. A summary of what each person needs help with, which is health information about people with disability. Patient data guide.

  4. 4
    A hold that reveals an incident

    A status that says a named man’s shift is under incident review. It tells a reader that something happened; the Incident Management Rules require the record of a reportable incident to be kept for 7 years.

  5. 5
    A dispute and a missing note

    That a mother is disputing a charge through a complaint, and that a shift has no note behind it. Both are about the service’s own records. The note is what backs the claim, and complaint information is to be kept confidential and disclosed only if required by law or otherwise appropriate. Privacy Act guide.

  6. 6
    Hours, rates and totals

    What the service charges and what each participant uses. Financial information about the participants’ plans and commercial information about the service. The rates in this sheet are invented.

  7. 7
    A formula that reads another column

    A sum that totals paid claims. It has no personal data in it, though in a real workbook a formula can pull values from other sheets and files that travel with the upload.

  8. 8
    The other tabs

    This workbook has tabs for plan managers, rates and notes. When a spreadsheet file is uploaded, everything in it goes: other sheets, hidden columns, cell comments, formulas that point at other files and properties such as the author.

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 service 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 participant 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 NDIS services.

Common questions

Do ChatGPT and Claude use the information in a claims and invoice spreadsheet?
Yes. Rows pasted into ChatGPT or Claude are sent to the AI company's servers and 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 lists each participant beside an NDIS number and the support received, and its status column shows a shift held under incident review and a charge disputed through a complaint.
Is it safe to paste an NDIS claims spreadsheet into ChatGPT or Claude?
No, because the sheet is a table of identifiers. 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. A column of NDIS numbers beside names and support types makes impersonation and claims fraud easier, and the status column tells a reader which shifts are under question. A worker who only wants a formula checked still sends every row pasted.
Can I upload the claims spreadsheet to ChatGPT or Claude instead of pasting it?
An uploaded workbook leaves the service as a whole file. It can carry other sheets, here plan managers, rates and notes, along with hidden columns, cell comments, tracked changes, formulas that point at other files and file properties such as the author. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. Pasting the visible rows is a disclosure, and an upload sends more than the rows on screen.
How can an NDIS provider use AI for a claims 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 answer is produced from the service's own claims records and the shift notes behind them. The service administrator checks it against the participant file before any claim is submitted.
What records does an NDIS provider keep to support a claim?
NDIA guidance on invoicing and record keeping expects providers to keep records of the supports they deliver, such as case notes, rosters and service agreements, so they can show a claim was for support actually provided. The claims sheet is the summary of those records. Retention periods for claim records are set elsewhere, so check the current NDIA terms rather than relying on this page.
Is an NDIS number sensitive?
It is personal information. The NDIS number identifies the participant to the NDIA and is used to match an NDIS provider's payment request to the participant's plan. Next to a name and a support type it makes impersonation and fraudulent claims easier. A column of NDIS numbers deserves the care a service gives a column of bank account numbers.
Why does a claims sheet carry more than it shows?
A workbook can hold hidden columns, other tabs, cell comments, formulas that point at other files and properties such as the author. Uploading the file to an AI tool sends all of it rather than just the tab on screen. Cleaning a copy first is slow and easy to get wrong, which is why the destination matters more than the clean-up.

See Dominion answering from your own documents.

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

Book a demo