AI and your data
What is sent to model providers and what is not, why your books are not training data, and how machine access is bounded and logged.
ReferenceFor Business owners, Firm staff, Portal clients, Bookkeepers
What is sent, and when
Model providers are used for specific tasks: reading a document you uploaded, proposing a category for a transaction, answering a question you asked. In each case what is sent is the material needed for that task — the document, the transaction and comparable history, or the question and the records that bear on it.
What is not sent:
- Your ledger in bulk. There is no wholesale export to a model provider.
- Data belonging to other organizations. Scope boundaries apply to machine access exactly as they do to human access.
- Credentials, API keys, or bank access tokens.
Training
How machine access is bounded
- Automation operates within one organization and the entity scope of the work it is doing.
- It cannot exceed the permissions of the context it runs in.
- Every action it takes is recorded in the audit trail with an actor and a timestamp.
- Approval policy and period locks constrain it exactly as they constrain a person.
Common questions
You can work entirely manually — enter documents by hand, categorize without proposals, and not use the coworker. Talk to us if you need automation disabled at the account level rather than simply unused.
That automation reads their documents and proposes coding, that a person reviews anything ambiguous, that every action is logged and attributable, and that their data is not training material. Those four points cover nearly every question a client actually has.
The AI disclosures page on the public site covers the formal position, alongside the privacy policy and the subprocessor list.
See alsoSecurity, access, and your data
Was this useful?