Pasting a contract or patient note straight into ChatGPT, Copilot, or Gemini is the fastest way to leak PII you cannot audit. Omit sits between copy and paste and redacts on the way in, so the model never sees the raw data.
| How they compare | Omit | pasting into AI |
|---|---|---|
| Where the sensitive text goes | Stays on your device. Only the redacted version is pasted | The full text is sent to a third-party AI server |
| PII exposure to the model | Names, emails, IBANs, and IDs are replaced before paste | Everything you paste is exposed, including PII you missed |
| Works offline | Yes. Redaction runs locally with no network | No. The chatbot requires a live connection |
| Audit trail | Per-run report of exactly what was redacted | None. You cannot prove what was sent |
| Setup effort | Install Omit Redact and enable Protect once | None. Pasting works in any browser with zero tooling |
| Deliberate raw override | Ctrl+Alt+V pastes raw text as a logged, on-purpose exception | Every paste is raw by default, with nothing logged |
| Rows where pasting into AI leads are marked plainly and explained below. We concede what we do not win. Verified 23 July 2026. | ||
Pasting is genuinely frictionless, and for text with no personal or client data it is perfectly fine. Omit is for the moments when the text does contain PII and a careless paste becomes a reportable incident. If you never handle sensitive data, you do not need us.
Published research on what employees actually paste into AI tools. Estimates vary widely because methodologies and definitions of sensitive differ, so the range is given rather than the most alarming single number.
Verified 23 July 2026.
Cyberhaven's analysis put roughly one in twenty employees as having pasted company data at least once. That was the early measurement and it is the conservative end of the range.
Cyberhaven analysisOne 2026 figure puts around 34.8 percent of employee prompts as containing sensitive company data, against 11 percent in 2023. Another study found sensitive information in over 4 percent of prompts but 20 percent of file uploads. Treat the exact number with caution: what counts as sensitive is defined differently in each study. The direction of travel is consistent across all of them.
Comparison of published figuresMost organisations already have an acceptable use policy covering AI tools. Policies are read once and the paste happens hundreds of times. A control that operates at the clipboard applies at the moment the decision is actually made, and it records the exception when someone deliberately overrides it.
How Protect worksChoose Omit when staff use AI tools with material you would not email to a stranger, and when you need to be able to show what was and was not exposed rather than assert that policy was followed.
If your team never handles personal or client data, pasting freely is fine and a redaction layer is friction you do not need. This page is not an argument against using AI tools. It is an argument about what goes into them.
Omit runs fully offline, and the Omit Redact beta is a free download. Read the Trust Center and verify the offline promise yourself.