AI Safety & Compliance

Last updated on 

Configure PII redaction, content guardrails, and usage limits for enterprise compliance. Essential for data protection and cost control.

Accessing AI Safety Settings

  1. Go to Settings in the sidebar
  2. Click the Account Settings tab (admin only)
  3. Find the AI Safety & Compliance card

PII Auto-Redaction

Automatically detect and mask personal information in AI responses.

Enabling PII Detection

Toggle Enable PII Detection & Redaction on.

Redaction Modes

Choose how PII is masked:

Mode Description Example
Full Redaction Replace with [REDACTED] [REDACTED]
Partial Masking Show last 4 characters ****1234
Hash for Analytics Deterministic hash for tracking a1b2c3d4

Partial Masking is recommended for most use cases.

What Gets Redacted

  • Credit card numbers
  • Phone numbers
  • Email addresses
  • Social Security numbers
  • Other PII patterns

Usage Limits

Control AI costs and prevent runaway spending.

Max Tokens per Conversation

  • Field: Number input
  • Purpose: Limit AI context window per conversation
  • Default: Leave empty for unlimited
  • Recommendation: Set to 4000 for cost control

Daily AI Budget (USD)

  • Field: Number input (e.g., 10.00)
  • Purpose: Cap daily AI spending
  • Default: Leave empty for unlimited
  • Usage: When budget is reached, AI pauses until next day

Recent Guardrail Events

Shows the last 5 safety events:

Column Description
Icon Severity indicator (red = critical, orange = high)
Violation Type What triggered the guardrail
Action Taken How the system responded
Timestamp When it occurred

This helps you monitor AI safety in real-time.

Best Practices

  1. Enable PII redaction for compliance: Required for GDPR, CCPA, HIPAA
  2. Set daily budgets: Prevent unexpected costs
  3. Review guardrail events weekly: Identify patterns
  4. Use partial masking: Allows verification while protecting data