AI and content protection notice

What is transmitted

The selected model may use OpenAI API or Anthropic API. Answers need your message, relevant history, summaries and tool results. Private text is pseudonymized on Prawai's servers. Original PDFs, images, Word files and restoration dictionaries are not sent to those models. “Local” here means Prawai infrastructure, not only your own device.

Review your materials and output

New attachments have a review step for detected data and corrections. Plain messages do not always open a review dialog but still pass through masking. OCR and detection can miss data: remove unnecessary information and check the text. Prawai restores known values for authorized recipients. Context may still make a person identifiable.

Responses can be wrong or incomplete. Check facts, quotations, dates, calculations, sources and drafts before use. Failure to verify a provision's applicability is not evidence that it does not apply. The model does not replace professional assessment.

Choice, withdrawal and provider handling

Enable or withdraw authorization in My account. A company workspace also requires a current authorized owner's choice. Declining preserves account and data-rights access. Withdrawal stops new transmissions and rechecked work, not transmissions already completed.

Retention and access depend on the provider, endpoint and organization configuration. A request not to store a response does not guarantee zero retention, for example in safety systems. We do not promise a special zero-retention agreement for this service. Entrusted data needs its own legal basis; AI authorization does not replace third-party consent or professional confidentiality requirements.