AI use at Mindkit
How does Mindkit use AI-assisted tools?
Mindkit uses approved AI-assisted tools for specific parts of clinical documentation and practice work. Different systems are approved for different kinds of information.
When an AI-assisted system produces written material, a clinician or staff member reviews it before it is used. The responsible clinician remains accountable for clinical interpretation, conclusions, recommendations and final clinical documents.
Mindkit does not put identifiable client information or clinical records into ChatGPT or other general-purpose consumer AI tools.
This policy explains which systems we use, what each system is allowed to receive, what happens during an appointment, and how information is handled.
Which AI-assisted tools does Mindkit use?
AI systems can have substantially different arrangements for encryption, storage, access, retention and model training. Mindkit decides which system can be used according to the information involved.
Heidi Health
Heidi Health is used during appointments to assist with clinical documentation. It processes the conversation, produces a written transcript and can generate a draft clinical note. The clinician reviews and edits the note before it becomes part of the clinical record.
BastionGPT
BastionGPT is the system Mindkit uses when AI assistance requires identifiable clinical information outside the Heidi documentation process.
It may be used to organise, summarise, draft or check material for clinical notes, correspondence and reports. The clinician remains responsible for interpreting the information and approving the final content.
Claude Enterprise
Mindkit uses a HIPAA-configured Claude Enterprise environment for work such as developing worksheets, handouts, information sheets, templates and other practice materials.
Claude Enterprise has different organisational privacy and security controls from consumer Claude. Mindkit applies an additional restriction to our use of it: identifiable client information is not entered into Claude Enterprise.
ChatGPT and other consumer AI
Mindkit does not enter clinical records, session transcripts, clinical notes, assessment reports, information from Halaxy, or other identifiable client information into ChatGPT or other general-purpose consumer AI tools.
These tools may be used for work that contains no identifiable client information.
What happens when Heidi is used during an appointment?
When Heidi is used during an appointment, it processes the conversation in real time to create a transcript from which a draft clinical note can be generated.
The audio recording is deleted immediately. Mindkit does not retain an audio recording of the appointment through Heidi.
Your clinician reviews and edits the draft note before it becomes part of your clinical record. The clinician decides what belongs in the note and corrects information that is inaccurate or incomplete.
Why is Heidi part of Mindkit’s standard service?
Heidi is part of Mindkit’s standard clinical documentation process. We do not offer a separate manual note-taking option.
Clinical documentation takes clinician time. Without Heidi, the clinician would need to spend more of the appointment taking notes or complete substantially more documentation after the appointment. Both affect how clinician time is allocated.
Heidi allows the clinician to concentrate on the appointment while the initial documentation is produced. It also reduces the administrative load attached to each clinical session.
Mindkit’s clinical team is neurodivergent. Reducing documentation load is one practical way we support clinicians to work sustainably and reduce the risk of burnout.
Documentation time is also part of the cost of providing a clinical service. Using Heidi allows us to include this work within our current fees while paying clinicians for the time involved.
Maintaining a separate manual documentation system for some clients would require clinicians to manage two clinical workflows across their caseload. Mindkit uses one standard documentation process so that the requirements remain consistent across the practice.
For these reasons, consent to Heidi is required to receive Mindkit clinical services
What information can AI tools receive?
The information an AI-assisted system can receive depends on which system is being used and what Mindkit has approved it to do
Identifiable information includes obvious details such as a person’s name, date of birth and contact information. A person can also be identifiable from a combination of details even when their name has been removed.
Clinical notes, transcripts, assessment material, reports and information taken from the clinical record are treated as sensitive health information.
Heidi receives information from the appointment because it is the approved system used for clinical documentation.
BastionGPT can receive identifiable clinical information when that information is required for the clinical task
Claude Enterprise is used without identifiable client information
General-purpose consumer AI tools are not used with identifiable client information or clinical records.
Removing a name is not enough if the remaining information could still identify the person. Mindkit’s internal instructions require staff to consider the information as a whole before using it outside an approved clinical system.
Does AI make clinical decisions?
No. A Mindkit clinician makes the clinical decision.
AI-assisted tools can help transcribe information, organise material and produce draft written content. They do not independently determine an Autism or ADHD assessment outcome, a neuropsychological conclusion, a treatment decision, a funding recommendation, or whether somebody can access a clinical service.
The clinician interprets the clinical information, develops the formulation and recommendations, and approves the final report or other clinical document.
AI-generated material is checked against the underlying information. Material that is inaccurate, incomplete or inconsistent with the clinician’s judgement is corrected or discarded.
What happens to information entered into these systems?
Security is assessed separately for each system. Encryption, where information is stored, who can access it, retention periods and model-training rules are different parts of that assessment.
Heidi Health
Heidi encrypts information while it is being transmitted and while it is stored. Heidi uses Australian data localisation for Australian customers.
Heidi states that consultation information, transcripts and generated clinical notes are not used to train its AI models.
The audio recording used during a Mindkit appointment is deleted immediately.
The written transcript is retained in Heidi for approximately one week while the relevant information is transferred into Mindkit’s clinical systems.
The final clinical note is stored as part of the client’s clinical record and is handled under Mindkit’s Privacy Policy and health-record obligations.
BastionGPT
Mindkit uses BastionGPT for AI-assisted work that requires identifiable health information.
BastionGPT uses TLS encryption while information is being transmitted and AES-256 encryption while information is stored. BastionGPT states that sensitive data from Australian customers is stored in Australia.
BastionGPT contractually states that customer information is not used to train AI models, sold, or used for marketing.
Temporary data used to process a request may be retained for up to 30 days. Information kept in the user account, including chats and uploaded documents, has separate retention and deletion controls.
Claude Enterprise
Mindkit uses a HIPAA-configured Claude Enterprise environment. Anthropic states that inputs and outputs from its commercial products are not used to train its models by default. Claude Enterprise also provides organisational controls over access and data retention.
Mindkit does not rely on those protections as permission to put clinical records into Claude. Identifiable client information is excluded from our use of Claude Enterprise.
Why are these details considered separately?
Encryption does not determine where information is stored or how long it remains there. Australian hosting does not by itself determine whether information can be used for model training.
Mindkit reviews the data-handling arrangements for each system and approves particular uses instead of treating every AI service as equivalent.
More detailed information about Mindkit’s handling of personal and health information is available in our Privacy Policy.
What if I have questions or do not want an AI-assisted tool used?
You can ask how a Mindkit AI-assisted system works, what information it receives and how that information is handled before you begin a service.
We provide information about Heidi before clinical services begin. Consent to Heidi is required because it is part of Mindkit’s standard clinical documentation process. We do not offer a separate manual note-taking option.
If you do not consent to Heidi, we cannot provide the clinical service. We can help you look for another provider whose documentation arrangements suit you.
If you have already started working with Mindkit and later withdraw your consent to Heidi, tell your clinician or the practice team. We will discuss what this means for your remaining appointments and can assist with transition to another provider where needed.
Questions about AI use or privacy can be sent to hello@mindkit.com.au.