Review input and output safety checks
Reviews & approvals · Example outcome
The handoff
A private submitted text, possible personal or secret data, and proposed public-facing responses review draft with source checks, uncertainty notes, and human approval status.
Keep in mind
A screening check may miss private information, harmful instructions, or unsafe output. Give clear notice, minimize personal data, secure secrets, test both false positives and missed cases, return a cautious fallback when uncertain, and require a qualified person to review flagged or consequential content.
