Why deleted documents can still appear in AI answers
Trace document deletion through chunks, embeddings, caches, jobs and conversation context with a synthetic matrix and explicit limits on removal claims.
Read the guideTopic 03
Sensitive data
Trace what leaves your app and what model providers receive.
What to do after sensitive data is sent in a model request
An operational response worksheet for accidental model-payload disclosure: stop repeat sends, map copies, preserve minimal evidence, and verify recovery without promising recall.
Read the guideRoll out pattern policies without turning exceptions into bypasses
Label synthetic false positives and missed cases, choose narrow responses, and bind temporary exceptions to one reviewed request and current policy.
Read the guideRender model output safely in HTML and Markdown
Define a model-output rendering policy for plain text, Markdown, links and remote images, then check it with a small synthetic fixture.
Read the guideSecure document uploads before AI preprocessing
A practical admission worksheet for uploaded documents: constrain formats and parser resources, separate remote fetching, and preserve authorization before indexing or model use.
Read the guideWhat sensitive data should an AI app send to a model?
Map model-request data, remove unnecessary fields, check endpoint retention, and test redaction with synthetic records.
Read the guide