What SafePrompt detects: keys, PII, secrets, and limits
End users often ask: “What exactly gets redacted?” This page lists the major categories, example shapes, and honest limits so you know what SafePrompt covers and what still needs your attention.
Category 1 — API keys
These usually become [API_KEY].
- AWS access key IDs:
AKIA…andASIA… - OpenAI-style keys:
sk-…(non-Stripe forms) - Anthropic:
sk-ant-… - Google API keys:
AIza… - GitHub tokens:
ghp_,gho_,ghs_, etc. - Slack tokens:
xoxb-,xoxp-, and similar - Stripe secrets:
sk_live_,sk_test_,rk_…
AKIAIOSFODNN7EXAMPLE,
ghp_abcdefghijklmnopqrstuvwxyz1234567890
Category 2 — PII
These usually become [PII].
- Email addresses:
name@example.com - Phone numbers: formats like
+1 (555) 123-4567 -
Credit cards: Visa/Mastercard-shaped numbers that also
pass Luhn checks (for example
4111111111111111)
In Balanced mode, some card-like numbers near invoice/payment words may be skipped to reduce false alarms. Strict mode is stricter.
Category 3 — Secrets
These usually become [SECRET].
- Private key blocks:
BEGIN PRIVATE KEY/ RSA / OpenSSH - Bearer tokens:
Bearer eyJ…style values -
AWS secret access keys: long random strings when AWS
context is nearby (labels like
aws_secret_access_key)
What SafePrompt may miss (be careful)
- Custom internal tokens with no known prefix
- Passwords that look like normal words
- Secrets split across many lines in unusual formats
- Images/screenshots (text inside images is not scanned)
- Secrets typed after you click Allow once
- New cloud key formats we have not added yet
False positives — when SafePrompt might over-block
- Long random IDs that look like tokens
- Sample keys in documentation (you can whitelist exact phrases)
- Some phone-like number strings in dense digit text
Fix options: Undo, whitelist the exact safe phrase, or disable one rule in Advanced settings.
Stats categories in the popup
- API Keys
- PII
- Secrets
Privacy reminder for detections
Counts are stored locally. SafePrompt does not keep a history of the actual secret strings for the dashboard.
Need a detector added?
Email safeprompt.help@gmail.com with the key type (no real secrets) and we can consider it for a future update.