AI outputs should be screened to detect and redact sensitive information before they are delivered to users. Techniques include:
- Regular expressions to identify restricted terms.
- Keyword filtering to prevent accidental leaks.
- LLM-based filters to catch nuanced disclosures.
