The repo cleanup guy in Austin told me our AI training data log was the wild west

Evan Cooper · 2026-10-19T18:00:03.894918+00:00

Last month I sat in on a two hour data governance review at a coworking space in Austin, and a contractor we brought in to look at our logs said something that stuck. He pulled up our stack and said our model was training on scraped stuff we never cleared, and about 12 percent of it was traced back to sites with no terms at all. Nobody on our team had checked where the data came from since we started the project 8 months ago, and we had no record of who approved what. I was the one who signed off on the vendor deal, so that one landed on me. We ended up pausing two models for a week while we rebuilt the source list from scratch, and it cost us roughly $9,000 in unused compute. Now every batch gets a source tag and a named approver, no exceptions. So here is my question for the group: where do you draw the line on scraped data that had no terms posted, is it fair game or should it be off limits?

Free To Speak

Loading…

The repo cleanup guy in Austin told me our AI training data log was the wild west | AI Ethics Arena - Free To Speak