Privacy redaction

Settings → PII redaction controls how Parlot handles personally identifiable information (PII) in conversational data. Redaction is on by default for new agents.

What redaction does

Parlot scrubs sensitive details such as names, addresses, dates of birth, phone numbers, emails, and secrets from stored transcripts and related content. Masked tokens (for example [PERSON] or [EMAIL]) replace identified spans so dashboards and judges can still read structure while identifiers stay masked.

When recordings are stored, Parlot also beeps out those sensitive spans in the audio so the sanitized recording matches the scrubbed transcript.

PII redaction is experimental. It can miss sensitive details or mis-label ordinary text. Treat redacted output as a best-effort aid. Before you export or share, review the content and confirm it meets your legal and policy requirements.

When redaction fails

If redaction cannot finish, Parlot withholds the transcript or recording. You may see messages such as “Content withheld — redaction unavailable” or “Recording withheld (PII redaction).” Only scrubbed or withheld content appears in the product.

Redaction still runs for closed sessions used in analytics. The setting’s critical effect is this withhold behavior when scrubbing fails or is unavailable.

Agent allowlists

Which agents get PII redaction uses the same New agents allowlist and Known agents overrides as recording, logs, and Gen AI content capture. See Agent policy allowlists for how those controls work and common setups. PII redaction is applied in Parlot from Settings only (no parlotize() override).

PII redaction defaults to on for new agents (* allowlist).

Where you see results

After a session closes and redaction finishes:

• Timeline and analysis views show scrubbed transcript text
• Evaluation judges consume redacted text
• Exports reflect the same protected content

See also Agent policy allowlists, Session logs and capture, and Recording.