Are customer data used to train AI models?
By default, no — your documents and case data are not used to train, fine-tune, or improve any AI model, whether operated by amaise or a subprocessor. Only if you explicitly opt in under a separate contractual agreement may your data be used to train or improve amaise's models — after personal data is removed (de-identified) — and those model improvements may benefit other customers as well.
This applies to amaise and all subprocessors alike:
Azure OpenAI (Microsoft): Contractually prohibited from using customer data for model training or improvement — regulated in the Microsoft Online Services DPA.
Other AI services: Data submitted for processing are not retained by the providers and not used for training.
amaise proprietary models: amaise operates its own specialized models, which are not trained on customer data by default.
This guarantee is documented in the data processing agreement (DPA) signed with every customer.
Evaluation datasets: To test and measure the quality of document processing, amaise can keep evaluation datasets built from customer documents. This is optional: a dataset is kept only if your organization has agreed to it in its contract with amaise (opt-in), and your organization can withdraw its consent (opt out) at any time. After a withdrawal, amaise takes no new data from your organization for evaluation datasets; datasets that are already reviewed and redacted are kept, and copies whose preparation is not finished are deleted.
Not part of contract processing: Evaluation datasets are not part of the processing under your service contract and never contain original documents.
Manually reviewed and redacted: Before a dataset is kept, the person preparing it manually reviews the data and removes and redacts personal data. The preparation is documented.
Not re-identifiable: The resulting data cannot be traced back to the original document owner or to the person concerned, such as an insured person or a patient.