OpenAI accidentally erased training findings as lawyers seek copyright violation

Strange.

I think this is saying that OpenAI provided two VMs, deleted one, restored from a half-recent backup, but then the restored data isn't usable in court. Why not? If it's about taint, then why could the data be usable in court before the imbroglio, if OpenAI had the practical ability to delete/change it?

  "accidentally"