Healthcare After-Hours Mental Health Crisis Call Transcripts: Use case and AI Confidentiality

Some clinics now answer the after-hours line with an AI call-taking tool that records the call, asks a set of scripted questions and passes the caller to the on-call clinician. When the caller is a patient in distress, the transcript holds the most sensitive words the clinic will ever store: how the person feels, what they have been drinking, the answers to a risk screen and the name of the person asleep in the next room. Staff paste it into ChatGPT or Claude to write the file note. This page marks every item in a transcript that the Surveillance Devices Act 2007 (NSW) and the Privacy Act bear on, with no method detail, and shows how the same transcript is handled on Ermos.

The after-hours mental health crisis call

Produced when a patient rings the after-hours line and the call is taken by an AI call-taking tool, which transcribes it and flags it for the on-call clinician. A clinician reviews the transcript, rings the patient back and records what was done.

After-hours call transcript

[00:00:03] Assistant:

[00:00:19] Caller:

[00:00:31] Assistant: Thank you. Can I take your name and date of birth?

[00:00:38] Caller:

[00:01:02] Caller:

[00:02:10] Assistant:

[00:02:21] Caller:

[00:02:40] Assistant:

[00:02:52] Caller:

[00:05:20] Assistant:

[00:06:00] Assistant:

[00:06:41] Assistant:

[00:06:50] Caller:

An invented transcript. No vendor is named and no method of self-harm is described. The mobile number is a placeholder. Hover or tap any highlighted item for the analysis; the full list is below.

Crisis call transcript summariser

An Agent+ that reads the transcript of an call to the night-time line and prepares a structured summary for the on-call clinician. It works from the practice's call protocol and risk screen form, and sets out what the caller said, what was asked and what is missing. It also drafts a call-back note.

What it does

  1. Reads the transcript and the practice's call protocol
  2. Summarises risk screen answers, safety plan and consents
  3. Flags questions not asked and details missing

What stays with your people. The on-call clinician reads the transcript, rings the patient back, makes every clinical decision and records what was done.

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 header: tool, line, date, time and length

    Identifies the clinic, the line, the exact call and the fact that an automated tool took it. A reader can match it to call logs and the roster for that night.

  2. 2
    The recording notice and the caller’s consent

    The tool says the call is recorded and asks for agreement, and the caller says yes. In NSW the Surveillance Devices Act 2007 s 7 prohibits knowingly using a listening device to record a private conversation to which the person is a party unless all principal parties consent (s 7(3)(a)) or one principal party consents and s 7(3)(b) applies. A conversation the parties would reasonably expect might be overheard is not private (s 4). The consent is to the recording; where the transcript is sent afterwards is a separate question. Privacy Act guide.

  3. 3
    Name and date of birth

    Two identifiers that tie the call to one person. Personal information collected in providing a health service, so health information (Privacy Act s 6FA(b)). Privacy Act guide.

  4. 4
    The caller’s account of his state, a drink and a partner

    A resignation, drinking tonight, a statement of hopelessness and the first name of a partner. Health information is sensitive information (s 6(1)), and it names a third party. Patient data guide.

  5. 5
    The risk questions and the answers

    The scripted questions asked and the caller’s replies, including that he had not acted or made a plan and who was with him. Among the most sensitive content a clinic holds. The clinical weight given to answers like these is a clinician’s judgement. Patient data guide.

  6. 6
    The safety steps agreed on the call

    What the caller was asked to do while waiting, drawn from his circumstances. Personal detail about where he is and who is near him. Patient data guide.

  7. 7
    Consent to speak to the partner

    The caller agrees that a clinician may speak to his partner. If a patient will not agree and the clinic believes a disclosure is needed, Privacy Act s 16A item 1 sets a permitted general situation: it must be unreasonable or impracticable to obtain consent (the OAIC says an entity should be able to point to clear reasons) and the entity must reasonably believe the disclosure is necessary to lessen or prevent a serious threat to the life, health or safety of any individual. The Act permits it in those conditions. It does not tell a clinician what to do. Privacy Act guide.

  8. 8
    Standard emergency wording

    Triple Zero and the Lifeline number, the same words the clinic gives every caller. Public information that identifies no one.

  9. 9
    The callback commitment and the caller’s number

    Who will ring, within what time, and on which mobile. The mobile is a direct contact for a named patient in distress. Privacy Act guide.

  10. 10
    The closing log line

    Times of escalation and the clinician’s return call, and the note that clinician review is needed before it is saved. Operational, but it records when the clinic acted that night.

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. See how this works in practice: Clinical documentation on Ermos.

Common questions

Do ChatGPT and Claude use the information in an after-hours crisis call transcript?
Pasting a transcript sends everything in it to the AI company's servers, where it is processed 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 transcript carries the caller's name and date of birth, his answers to the clinic's risk questions and the first name of his partner.
Is it safe to paste an after-hours crisis call transcript into ChatGPT or Claude?
No, because a call like this is among the most sensitive content a clinic holds. 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. It is health information, which is sensitive information under the Privacy Act. 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. Leaving out the name still leaves a date of birth, a call to the clinic's after-hours line at 9.42pm on a Saturday and a partner's first name.
What if an AI call-taking tool records and transcribes the after-hours call?
Recording and transcription tools send the audio or transcript to the tool's vendor, so the same questions apply. The practice needs to know where the transcript is stored, whether it trains a model and who can review it. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. In this transcript the caller agreed to the recording, but that consent covers the recording and says nothing about where the transcript goes afterwards.
How can a medical practice use AI for a crisis call transcript 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 file note is drafted from the call record and the practice's own after-hours template. The on-call doctor checks it against what was said and signs it, and clinician review happens before anything reaches the patient record.
Can a practice record an after-hours call in NSW?
The Surveillance Devices Act 2007 (NSW) s 7 prohibits knowingly using a listening device to record a private conversation to which the person is a party unless all principal parties consent, or one consents and a further condition in s 7(3)(b) is met. A recorded notice that asks the caller to agree is one common approach. Whether a call is private depends on the circumstances, so a practice should take legal advice.
When can a practice pass on a caller's details to prevent serious harm?
Privacy Act s 16A item 1 describes a permitted general situation where it is unreasonable or impracticable to obtain the person's consent and the entity reasonably believes the disclosure is necessary to lessen or prevent a serious threat to the life, health or safety of any individual. APP 6.2(c) allows a use or disclosure in that situation. The Act permits it and does not direct what a clinician should do.
Is a transcript of a crisis call health information?
Yes. Section 6FA of the Privacy Act counts information about a person's health, and other personal information collected in providing a health service, as health information, and section 6(1) lists health information as sensitive information. A transcript that holds a name, a state of mind and answers to a risk screen is health information, whether or not it has been saved to the record.

See Dominion answering from your own documents.

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

Book a demo