Sensitive Information Disclosure
The model reveals data it shouldn’t — PII, secrets, or another user’s records — in its answer or by over-fetching from connected systems.
What it means
Sensitive information disclosure is when confidential data ends up in a model’s response. It can be personal data, credentials, internal business information, or one user seeing another user’s records.
It often happens by accident rather than attack: the model is given broader access to data than the current user should see, and simply surfaces what it was handed.
Why it matters
Disclosure carries direct privacy and regulatory consequences and erodes user trust quickly.
In data-connected assistants the usual cause is over-broad access — the model can read fields or records that the person in front of it was never entitled to.
How we approach it
Default-deny field exposure: for every endpoint, only the fields you explicitly allow are ever shown to the model. A newly added field stays hidden until you opt it in, so data can’t leak into a prompt just because it appeared in an API response.
Sensitive values are masked before they reach the model — email addresses, phone numbers, and card-style numbers are redacted rather than passed through in the clear.
Scoped identity: the assistant acts within the signed-in user’s own identity, so it sees what that user may see and nothing filed under someone else.
Credentials never travel to the browser — model-provider keys are stored encrypted and are never returned to the client.
Frequently asked
Is my data used to train the model?
No. Your data is used to answer the question at hand and is not used to train or fine-tune any model.
Which fields can the assistant see?
Only the fields you explicitly allow for each endpoint. The default is to hide everything else, including fields added later, until you choose to expose them.
How are provider API keys protected?
They are stored encrypted and are never sent to the browser. Responses expose only whether a key is configured, not the key itself.