Security & privacy · 2026-10-17
Does MailTrace train AI on your private emails?
AI processing is not the same as AI training. See what MailTrace says about user emails, case corrections, external providers and retention.
Asking an AI question does not normally retrain it
Suppose you ask an assistant, 'Which email first mentioned the delayed payment?' The system can search relevant messages and use a model to help compose an answer. That is a use of an existing model, called inference.
Training is different: the model's internal parameters are updated using examples. Fine-tuning is a specialised form of additional training. Neither is a necessary step in answering an ordinary email question.
What the MailTrace privacy policy says
MailTrace states that it does not use users' emails, attachments, case reports or questions to train or fine-tune a shared internal AI model. It also says that saved corrections help organise evidence in your own case rather than train a model shared across customers.
Those statements concern MailTrace's own use of data. They should not be confused with the separate contractual or technical terms of an external AI service that you approve for processing.
What about an outside language model?
When you approve an AI-assisted operation, the configured external provider may receive a question and relevant evidence needed for that task. Providers differ in how they retain requests, handle safety logs and offer optional data controls.
For that reason, 'we do not train on your emails' should not be rewritten as 'nobody ever stores or sees your emails'. A trustworthy privacy explanation makes both the processing pathway and applicable provider terms clear.
A simple example
Imagine three fictional messages: one promises a refund, the next changes its date and the third confirms payment. The model can be asked to summarise them and link each finding to the right message. That summary can be generated without updating the model itself.
The result still deserves a human check: was the promised date quoted correctly, and does the linked email support the finding?
The takeaway
Using AI, storing evidence and training a model are separate processes. MailTrace's published no-internal-training policy is useful, but decisions about external AI processing and retention deserve equal attention.