Ask HN: AI data classification – Bigger headache than protecting it?

Hey HN, We're all rushing to use AI/LLMs with internal data. But before protecting it (masking, etc.), how are you figuring out what's actually sensitive (PII, IP) for AI use?

1. Current approach to classifying this data for AI? (Manual, scripts, tools?) 2. Biggest pain points/time sinks in that classification step? 3-Is this classification challenge largely solved, or still a manual mess for most? 4-Magic wand: What would your ideal 'AI data classifier' do? Curious about your real-world trenches, not generic AI risk talk. What works/doesn't? Cheers!