> Provenance is hard to track - you would hope OpenAI has very good tools for this, but a full data trail of all inputs is difficult to trace through.

What would OpenAIs incentive for this be? They've gotten away with scraping everything and getting it ruled fair use. It seems like willful ignorance is an affirmative defense today. Why would they want to have some sort of audit trail that could prove otherwise?