Pathology result emails and AI data leakage: what an abnormal result email gives away in ChatGPT or Claude.

When a result comes back abnormal, the GP often emails the practice nurse or reception to get the patient in quickly, and the email is written fast. It names the patient, quotes the values, says what the doctor suspects and tells staff how to make the call. Staff and doctors paste these emails and the reports behind them into ChatGPT or Claude to summarise them or draft the recall. This page marks every item in a routine abnormal-result email that the Privacy Act, the Notifiable Data Breaches scheme and Good medical practice bear on, and shows how the same email is handled on Ermos inside the practice.

The internal abnormal pathology result email

Sent by a GP who has reviewed an abnormal result and needs staff to contact and book the patient. Common when the doctor is about to be away, so the instruction has to stand on its own.

An invented email. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the practice

1. 1
   The patient, in the subject line

   Name, date of birth and the headline values in the one place most likely to be forwarded or pasted. Health information once it is tied to a name. Patient data guide.

2. 2
   The laboratory values

   Haemoglobin, ferritin, MCV and a positive faecal test against a named patient. Health information is sensitive information under the Privacy Act. Patient data guide.

3. 3
   The interpretation and the family history

   The doctor’s reading that this may be cancer, with a parent’s cancer as the reason. Clinical judgement, and the most alarming sentence in the patient’s record at that point. Patient data guide.

4. 4
   How to make the call, and who must not be told

   An instruction about voicemail and the patient’s wife. It records the patient’s privacy risk and the practice’s plan for it, and a leak could undo the care taken. Privacy Act guide.

5. 5
   The doctor’s leave and the cover

   Who is away and who covers. Operational detail, but it shows the practice’s staffing on a particular week.

6. 6
   An attachment named for the patient

   A file name that carries the patient’s surname and initial, and a report and note behind it. An email thread leaks its attachments and their names too. Patient data guide.

7. 7
   The results policy

   The practice’s standard rule: two phone attempts, then a letter. Generic wording that identifies no patient.

8. 8
   The sender and the sign-off

   Ties the email to a named doctor. A doctor’s name is public on the register, but it identifies the source of the leak.

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

Part of AI data leakage in healthcare practices. See how this works in practice: Correspondence summaries on Ermos.

Common questions