To AI interception
What an employee sees when sensitive data is about to leave for an AI tool. By default Zeflin redacts the sensitive span and sends the cleaned prompt (no blocked work); whole tools can be blocked, and individual categories can be set to warn.
Sensitive data redacted before sending
To AI · a scrubbed version was sent
Zeflin removed the sensitive parts of your message before it left your device — ChatGPT received a scrubbed version, so you still get an answer. Its reply may refer to [REDACTED].
…Acct [REDACTED] and the API key [REDACTED] included below…
Zeflin names the category only — the raw value never left your device and isn't shown here.
Enforcement is outbound-only — redact by default, no per-prompt block.
Only data heading into an AI tool is enforced: by default Zeflin redacts the sensitive span and sends the cleaned prompt (never blocking the employee's work); a category can be set to warn, and a whole tool can be blocked — a deliberate admin opt-in. The reverse direction (AI content landing in real work) is monitor, warn, and attest only; it is never blocked.