Family emails and AI confidentiality: what an update after an NDIS incident gives away in ChatGPT or Claude.

Writing to a participant’s family after an incident is the hardest writing a service manager does. It has to be accurate, kind and careful about what can be said. That is why managers ask ChatGPT or Claude to help. The email carries the participant’s injury, the hospital, the worker involved, whatever the service has to admit and something the participant asked to keep from his own daughter. This page marks every item in an email to a nominee after a fall that the Practice Standards, the Complaints Rules and the Privacy Act bear on, and shows how the same email is drafted on Ermos inside the service.

The email to a participant’s family

Sent by the service manager to the participant’s nominee or family contact, with the participant’s consent where he can give it, after an incident. The Information Management standard expects consent to cover the purpose of a disclosure and when information may be disclosed.

To:
From:
Date: Wednesday 21 October 2026, 5.12pm
Subject: : update after yesterday

Dear Karen,

Thank you for taking my call yesterday. I want to put in writing what we know.

Warm regards,
Meena

An invented email. The email address is a reserved example address. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the service

  1. 1
    The recipient and the participant

    A named nominee at an email address, and the man the email is about. Personal information, and the link between a family and a service. Privacy Act guide.

  2. 2
    The injury

    What happened, which bone was broken and that he was conscious. Health information, which is sensitive information under the Privacy Act. Patient data guide.

  3. 3
    The hospital and the discharge plan

    Where he is, what treatment is planned and when he is expected home. Health information, and a statement of when he will be alone. Patient data guide.

  4. 4
    The worker and the notification

    That a worker has been taken off the roster and that the Commission has been told. An allegation that is not a finding. The Rules treat alleged incidents as reportable, and the Complaints Rules require procedural fairness when a complaint is dealt with. Privacy Act guide.

  5. 5
    An admission

    A written admission that a safeguard was recommended and not put in place. The manager’s decision, and the sentence that matters most if the matter is later disputed.

  6. 6
    A request about his daughter

    What Graham has asked that his daughter not be told and why. A confidence given by a participant about his own family, which the Code of Conduct requires workers to respect. Privacy Act guide.

  7. 7
    How to complain

    The complaint avenues. The Complaints Rules require reasonable steps to ensure a complainant and each affected person with disability are advised how a complaint can be raised with the Commissioner. Wording the service uses in every response.

  8. 8
    The changes and the meeting

    A change in how Graham is supported, from a named day, and an appointment at his home. Operational detail that shows when and where.

  9. 9
    The sender

    Ties the email and its admissions to the responsible manager.

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 an email to a participant's family?
Yes. A draft pasted in for polishing is sent to the AI company's servers and processed there to produce the 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 email to Karen carries Graham's broken wrist, the hospital and discharge plan, a worker taken off the roster and a request not to tell his daughter.
Is it safe to paste a family email into ChatGPT or Claude?
No, because the email holds health information, an allegation and an admission together. 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. Graham's injury and treatment are health information, the worker taken off the roster is an allegation that is not yet a finding, and the sentence admitting the roster had not changed is one the manager should choose to write.
Can ChatGPT or Claude be trusted to draft an email to a participant's family?
Only with a careful check. Language models can produce fluent text with details that are not in the source, wrong dates or references that look real but are not. A qualified person should check every item against the file. Here the injury, the discharge date, what the Commission has been told, the admission about the roster and the request about Graham's daughter would do the most harm if wrong.
How can an NDIS provider use AI for a family email 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 draft is produced from the incident record on the participant file and the service's own email template. The service manager decides what is said about the worker and any admission, checks every fact against the file and sends it.
Can a service tell a participant's sister about a fall?
It depends on who the participant has agreed may be told and whether she is a nominee. The Information Management standard expects consent to disclose information, including the purpose and when disclosure may occur without consent. A nominee or family contact named in the support plan is the usual basis. A private request from the participant about another relative should be respected and recorded.
What should a response to a family say about how to complain?
The Complaints Management and Resolution Rules require an NDIS provider's system to require reasonable steps to ensure that a complainant, and each person with disability affected by the complaint, are advised how it can be raised with the Commissioner, and that appropriate support in contacting the Commissioner is provided. Many services put a short standard paragraph in every response.
Is it appropriate to name the worker in an email to the family?
A worker is entitled to fairness, and an allegation is not a finding. The Incident Management Rules treat alleged incidents as reportable incidents, and the Complaints Rules require procedural fairness when a complaint is dealt with. Many services say only that a worker has been taken off the roster during a review. What is said is the manager's decision and should be made before a draft is written.

See Dominion answering from your own documents.

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

Book a demo