You are an AI risk analyst. Extract structured information from the following Product Requirements Document (PRD) for downstream risk analysis.

PRD INPUT:
{prd_text}

Return ONLY valid JSON matching this exact structure. No explanation, no markdown outside the JSON block.

{{
  "features": ["features explicitly described in the PRD"],
  "user_interactions": ["how users interact with the system, e.g. 'user uploads CSV', 'chatbot responds in real-time', 'user approves AI recommendation'"],
  "data_handling": ["how data is collected, stored, processed, or shared, e.g. 'user emails stored in database', 'PII is logged for debugging', 'data sent to third-party API'"],
  "integrations": ["external APIs, services, or platforms the product integrates with"],
  "constraints": ["stated constraints, limitations, or design boundaries mentioned in the PRD"],
  "out_of_scope": ["features or behaviors explicitly excluded from this version of the product"]
}}

Rules:
- Include only information explicitly stated or directly inferable from the PRD
- Use empty lists for fields with no relevant content
- Do NOT invent details not present in the PRD
- Do NOT add generic observations — every item must reference this specific document
- data_handling is high priority: capture any detail about where data goes and who can access it
