Redact Sensitive Information Before It Enters Your Twin's Dataset

ROLE You are a privacy-conscious dataset curator. OBJECTIVE Prepare source material for an AI twin without exposing information it does not need. CONTEXT The source material is [SOURCE MATERIAL]. The twin's purpose is [TWIN PURPOSE]. Relevant confidentiality requirements are [CONFIDENTIALITY REQUIREMENTS]. Details that must remain useful are [ESSENTIAL CONTEXT]. INSTRUCTIONS First identify direct identifiers, indirect identifiers, confidential business information, and details whose combination could identify someone. Recommend whether each category should be removed, generalized, replaced consistently, or excluded entirely. Show a sanitized version of a representative excerpt and explain which learning value it preserves. Create a transformation log that describes changes without repeating the sensitive content. Identify examples where redaction would destroy the lesson and recommend a separately written synthetic example instead. Distinguish anonymization claims from simple pseudonymization. CONSTRAINTS Do not guarantee anonymity. Do not reproduce secrets in the explanation. Avoid fabricated statistics and flag assumptions. Treat uncertain permissions as a reason to pause ingestion, not as implied consent. OUTPUT FORMAT Risk inventory, transformation rules, sanitized excerpt, transformation log, and review checklist.