Healthcare Adverse Outcome Letters to Patients: Use case and AI Confidentiality

When a procedure goes wrong, the patient is owed an honest account, an apology and a plan. The letter that follows the first conversation is among the hardest a clinic writes, so doctors and managers ask ChatGPT or Claude for a first draft and paste in the clinical timeline, the internal findings and the plan for the review. The result names the patient, describes the harm, states what the clinic got wrong and says the clinic’s insurer has been told. This page marks every item in a routine open disclosure letter that the Privacy Act, the Civil Liability Act 2002 (NSW) and the Australian Open Disclosure Framework bear on, and shows how the same letter is drafted on Ermos.

The letter to a patient after an adverse outcome

Sent after the first open disclosure conversation with a patient, by the clinician leading the disclosure, to record what was said and what the clinic will do. The Australian Open Disclosure Framework says each discussion should be documented and a copy provided to the patient.

Kestrel Lane Medical Practice
14 Kestrel Lane, Wollstonecraft NSW 2065 · Tuesday 10 November 2026

Dear Mr Mackintosh,

What happened.

What we know went wrong.

I am sorry.

The review.

What we have already changed.

Meeting and support.

Yours sincerely,
Dr Helena Voss, Kestrel Lane Medical Practice

An invented letter. Phone numbers are placeholders. Hover or tap any highlighted item for the analysis; the full list is below.

Adverse outcome letter drafter

An Agent+ that drafts the letter recording what was said in the first open disclosure conversation. It works from the clinician's notes, the clinical record and your own letter template, and sets out the apology, the facts as recorded and the review that follows.

What it does

  1. Reads the disclosure notes and the relevant record entries
  2. Drafts the letter in your template wording
  3. Flags facts the record does not yet support

What stays with your people. The clinician who led the disclosure reads, settles and sends the letter. The practice manager checks the insurer note. The agent drafts only.

An Agent+ is built once and anyone in the practice can run it. It works from your own documents and runs on Ermos Dominion, the AI unit in your office, or on Ermos Edge, your own instance in an Australian data centre.

What leaves the practice

  1. 1
    The patient’s name and address

    A named patient and a home address in a letter about his care. Personal information that is also health information because it was collected in providing a health service (Privacy Act s 6FA(b)). Privacy Act guide.

  2. 2
    The clinical timeline

    The procedure, the date, the symptoms, the hospital stay and the treatment. Health information is sensitive information (s 6(1)), and the timeline also sets out the practice’s own account of the harm. Patient data guide.

  3. 3
    What the clinic says it got wrong

    Plain statements that a sheet was not given and a review was not booked. The Framework asks for a factual explanation, without speculation and without apportioning blame. The wording of facts is a decision for the clinician, and the practice may want advice on it before the letter goes. Patient data guide.

  4. 4
    The apology

    The Framework (June 2026) says an apology or expression of regret should use the words I am sorry or we are sorry. In NSW, Civil Liability Act 2002 s 68 defines an apology as an expression of sympathy or regret, or of a general sense of benevolence or compassion, whether or not it admits fault, and s 69 provides that an apology is not an express or implied admission of fault or liability, is not relevant to determining fault or liability, and is not admissible in civil proceedings as evidence of fault or liability. Part 10 does not apply to liability excluded by s 3B or to defamation (s 67). Whether a particular sentence is part of an apology is a question for legal advice. Patient data guide.

  5. 5
    What the clinic does not yet know

    An honest statement that the cause and effect are unknown. The Framework says not to speculate on causes and not to pre-empt the results of reviews. A tool asked to improve this sentence may make it more certain than the facts allow. Patient data guide.

  6. 6
    The review: who, what and when

    Names the person leading the review, the scope, who is left out of it and the date it should finish. Internal governance that the practice would normally control, and a promise the practice is held to.

  7. 7
    Changes already made

    Process fixes that follow the incident. They describe a weakness and the clinic’s response to it, and the Framework expects updates on actions to prevent recurrence to be shared with the patient.

  8. 8
    A meeting, a contact and an offer

    A named contact with a direct line, an offer of independent support, and that care for the infection will not be charged. A commercial gesture to one patient, and a way to reach the practice manager.

  9. 9
    The insurer notice

    That the clinic has told its professional indemnity insurer. Sensitive to the practice, and a matter on which it may want advice about how to describe it.

  10. 10
    Where to complain next

    The Health Care Complaints Commission deals with complaints about health service providers in NSW. Public information that identifies no patient.

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

Ermos Dominion

The document is drafted on a unit in your office, from your own templates and the patient record. 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 Use cases in healthcare practices.

Common questions

Do ChatGPT and Claude use the information in a letter to a patient after an adverse outcome?
Everything in a pasted draft is 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. This letter carries Mr Mackintosh's name and home address, the timeline of his wound infection and hospital stay, and the practice's plain statement of what it did not do.
Can we use ChatGPT or Claude for a letter after an adverse outcome if we remove the names?
Not safely, because the story still points to one man. 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. A skin lesion removed on 15 October, an infection found on 4 November and a stay at Royal North Shore Hospital until 6 November are enough for his family and colleagues to recognise him. The timeline is health information, which is sensitive information. 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.
Does a ChatGPT Business or Claude Team plan make it safe for a letter after an adverse outcome?
Not by itself. A business plan changes the default on training, but it does not change who receives the content. On a personal plan the user cannot choose where conversations are stored or processed. Some business plans let an organisation choose to store data at rest in Australia (OpenAI), which still leaves the content with the AI company. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. The letter would still reach the AI company with the address, the timeline, the review plan and the mention of the insurer. Our comparisons with ChatGPT Enterprise and Claude weigh the trade-offs.
How can a medical practice use AI for a letter after an adverse outcome 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 letter is drafted from the clinical record and the practice's own letter template. The treating doctor checks every fact, decides what the practice states and takes advice on the wording where needed before signing.
Is an apology to a patient an admission of liability in NSW?
Civil Liability Act 2002 (NSW) s 69 provides that an apology is not an express or implied admission of fault or liability and is not relevant to determining it, and evidence of it is not admissible in civil proceedings as evidence of fault or liability. Part 10 does not apply to liability excluded by s 3B or to defamation. A practice should take legal advice on individual cases.
What should a letter after an adverse outcome contain?
The Australian Open Disclosure Framework (June 2026) says discussions should be documented with a copy to the patient, and follow-up should give clear information about further care, updates on reviews and actions to prevent recurrence, and the contact details of a nominated person. It asks for facts without speculation, blame or pre-empting the review.
Can an AI tool draft the letter?
It can help with tone and structure, but a person must check every fact and decide what the practice states. The Framework asks for plain language, no speculation and no pre-empting of review results. Where the draft is written also matters, because the timeline, the clinical detail and the practice's own findings go in with the request.

See Dominion answering from your own documents.

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

Book a demo