Can clinicians use ChatGPT with patient information?
Pasting identifiable patient details into ChatGPT is very likely a disclosure under the Privacy Act. What the OAIC and Ahpra say, and safer options.
8 min read
In Australia, putting identifiable patient information into a public generative AI tool such as ChatGPT is very likely a disclosure of sensitive health information. The Office of the Australian Information Commissioner (OAIC) recommends, as best practice, that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools. Ahpra's guidance warns that personal information entered into public tools may become public. Clinicians can use AI with patient information, but only through tools set up for clinical use, with the family's consent and a clinician reviewing every output.
This is general information, not legal advice.
Is it legal to put patient information into ChatGPT?
In most cases it would not comply with the Privacy Act, unless consent or another exception applies. The OAIC's guidance on privacy and commercially available AI products says privacy obligations apply to any personal information entered into an AI system, and to AI output that contains personal information. It also says that entering personal information into a public chatbot is likely a disclosure to the tool's owner, and that a disclosure needs a basis under Australian Privacy Principle (APP) 6.
APP 6 says you may use or disclose personal information only for the purpose you collected it for, unless the person consents or a listed exception applies. For sensitive information such as health information, the secondary purpose must be directly related to the primary purpose and within the person's reasonable expectations. A patient who gave you information for therapy would rarely expect it to go to a chatbot operator, so most clinical uses would rely on consent. Consent must be adequately informed, voluntary, current and specific, and given by someone with capacity.
So the careful reading is that pasting identifiable patient information into a public tool is very likely to breach APP 6 unless consent or another exception applies. It also conflicts with the OAIC's best-practice advice, which is separate from the question of what the law strictly requires.
Why health information is treated differently
Under the Privacy Act, health information is sensitive information, a subset of personal information with stricter rules. Personal information is information or an opinion about an identified individual, or an individual who is reasonably identifiable, whether true or not.
The small business exemption (the $3 million turnover threshold) does not apply to an organisation that provides a health service and holds health information, so every private practice is covered regardless of size. The OAIC's page on health service providers explains who counts.
The OAIC describes disclosure as making information accessible to others outside your organisation and releasing it from your effective control. Once text is submitted to a public chatbot, you cannot say what happens to it.
What the OAIC says about public AI tools
The same guidance says AI-generated or inferred personal information, including wrong or fabricated information, is a collection under APP 3. It asks for human oversight and ongoing monitoring of outputs (APP 10), updated privacy policies and notices that explain AI use (APP 1 and APP 5), and a privacy impact assessment. Due diligence on an AI product is a continuing task.
What Ahpra and professional bodies expect
Ahpra's page on meeting your professional obligations when using AI in healthcare applies the existing National Board codes of conduct to AI. Practitioners must apply human judgement to any output, understand where a product stores data and whether it retrains on it, and tell patients about AI use. Informed consent is generally needed when AI uses their personal data, and Ahpra suggests noting the patient's response in the health record. Ahpra also warns that personal information entered into a public generative AI tool may become public.
Ahpra's case studies include one about a practitioner who dictates findings into ChatGPT to tidy them up. Its lessons cover consent, offshore storage, developer use of the data, misread clinical terms and verifying the output.
Occupational therapists, psychologists and medical practitioners, including paediatricians, are Ahpra-registered, so this applies to them directly.
Speech pathologists are not. Speech pathology is self-regulated through Speech Pathology Australia (SPA) and Certified Practising Speech Pathologist (CPSP) status, as SPA's regulation fact sheet explains. SPA's position statement, Artificial intelligence in speech pathology: Ethical considerations (May 2024), supports AI that contributes to safe and effective services but requires informed consent, confidentiality and accountability. AI should not replace the speech pathologist's professional decision making, and AI-generated information needs detailed critique for currency, accuracy and relevance.
Psychologists have further guidance. According to an Australian Psychological Society (APS) article on its new practice guidelines (the full text is members-only), psychologists keep full responsibility for AI outputs. Written informed consent is expected before AI touches client data, and the consent should be specific to each tool, since blanket consent for "AI use" is not sufficient.
Does removing the name make it safe?
Often it does not. Names are only one way to identify a person, and personal information includes information about someone who is reasonably identifiable. In paediatrics, age, school, suburb, a rare condition, appointment dates and family details can point to one child with the name removed.
The OAIC describes de-identification as risk management rather than an exact science: removing direct identifiers is the first step, and other identifying information must also be removed or altered until the risk of re-identification is very low. Deleting a name rarely meets that standard. Our article on de-identification for Australian clinicians covers this in more detail.
A worked example
Priya (a fictional speech pathologist) has just finished a session with Max, a fictional 6-year-old. Her notes mention that he attends a named primary school, has a diagnosis of developmental language disorder, and that his mother has asked about support in the classroom. To save time, she pastes the summary into a public chatbot and asks it to tidy the wording.
The privacy issues are these:
- The summary is health information about a child who is reasonably identifiable from the school, diagnosis and family details, even without his name.
- Pasting it in is very likely a disclosure to the tool's owner, which needs an APP 6 basis. Max's parents were not told and did not consent.
- Priya cannot say where the text is stored, whether it leaves Australia (APP 8), or whether it trains a model.
- The tidied version may contain errors or invented detail (APP 10).
The alternative is a tool set up for clinical use with a documented privacy basis. Max's parents are told and agree, the tool has clear answers on storage and access, and Priya reviews, edits and signs whatever it drafts. If no such tool is available, she tidies the notes herself.
What a safer setup looks like
Before using any AI tool with patient information, ask the vendor, and record the answers:
- Where is data stored, and does it leave Australia (APP 8)?
- Are inputs used to train or improve models?
- Who can access the data, including the vendor's staff and sub-processors?
- How long is data kept, and how is it deleted on request?
- How does the tool support a consent workflow, so you can tell families and record their response?
- How does the tool require human review and sign-off before anything enters the record?
- What can the vendor tell you for your privacy policy? The automated decision-making transparency changes take effect on 10 December 2026, and the OAIC says a decision can be in scope even when a human reviews the output. Vendors should give clear high-level information so you can describe the personal information used and the kinds of decisions involved.
Business or enterprise versions of public tools can have different terms from free consumer versions. Read the actual terms and take advice before relying on them.
What to do if it has already happened
- Stop using the tool for patient information.
- Record what was entered, when, which tool and account, and who was affected.
- Assess the situation under the OAIC's Notifiable Data Breaches scheme, which requires eligible breaches likely to cause serious harm to be notified to affected people and the OAIC. Whether this incident is an eligible breach depends on the facts, so check the OAIC's current guidance.
- Get advice from your practice's privacy officer, your professional body, your indemnity insurer or a privacy lawyer.
- Review your practice's AI policy so it does not happen again.
In short
- Entering identifiable patient information into a public generative AI tool is very likely a disclosure that needs an APP 6 basis, and it conflicts with the OAIC's best-practice advice.
- The small business exemption does not apply to a practice that provides a health service and holds health information.
- Removing a name rarely makes a paediatric summary safe, because school, suburb, dates and diagnosis can still identify a child.
- Use AI for patient information only through tools set up for clinical use, with specific consent, a privacy basis and human review before anything enters the record.
- If it has already happened, stop, record it, assess it under the Notifiable Data Breaches scheme and get advice.
FAQ
Can I use ChatGPT for clinical notes in Australia?
Not with identifiable patient information in a public version of the tool. Doing so is very likely a disclosure that needs an APP 6 basis, and the OAIC recommends as best practice that personal and particularly sensitive information is not entered into publicly available generative AI tools. A tool set up for clinical use, with consent and clinician review, is the safer route.
Is ChatGPT compliant with the Privacy Act?
Your Privacy Act obligations apply to personal information you enter into any AI system. Whether a particular use complies depends on whether you have an APP 6 basis, how the data is handled and whether the other principles are met.
Does Ahpra say clinicians cannot use generative AI?
No. Ahpra explains how existing codes of conduct apply to AI use. It expects human judgement on outputs, an understanding of how a product uses data, and informed consent, and it warns that information entered into public tools may become public.
Is it still a problem if I remove the child's name first?
It can be. A school, suburb, rare diagnosis, dates or family details can identify a child without a name. Whether removing identifiers is enough depends on context, and the OAIC treats de-identification as risk management.
Syntropic Health drafts reports inside the case, and the clinician reviews, edits and signs every one. Its ambient scribe drafts notes on the same terms, for the clinician to check before signing.
Sources
Get new resources by email
New research and practical guides for paediatric practice.
Share this article
More resources
Take the next step
Book a demo, or explore the platform, trust page and research lab.