[matchagaucho]

This is actually a good classification problem for GPT/LLM.

Provide a range of leakage-seeking prompts and assign:

  IsLeakage: true/false