PII Filtering

Autrace scans every prompt and response for Personally Identifiable Information before it reaches the LLM or is stored in logs.

What's detected

Type	Example	Stored as
Email	user@domain.com	[EMAIL_REDACTED]
US SSN	123-45-6789	[SSN_REDACTED]
Credit card	4111 1111 1111 1111	[CREDIT_CARD_REDACTED]
Phone number	+1 (555) 123-4567	[PHONE_REDACTED]
IP address	192.168.1.1	[IP_REDACTED]
AWS key	AKIA...	[AWS_KEY_REDACTED]

Default behavior

PII is redacted (replaced with a placeholder) before forwarding to the LLM and before storing in the audit log.

The original prompt is never stored.

Configuration

In Dashboard → Rules, you can create PII rules to change the action:

Action	What happens
redact	Replace with placeholder (default)
block	Reject the request with 403
log	Pass through but log the detection

Testing

curl https://gateway.autraceai.com/v1/chat/completions \
  -H "Authorization: Bearer aut_live_YOUR_KEY" \
  -d '{"model":"openai/gpt-5.5","messages":[{"role":"user","content":"My SSN is 123-45-6789"}]}'

Check Dashboard → Logs — the stored prompt shows [SSN_REDACTED].