Comments - Delete Doesn't Mean Deleted. Just Ask OpenAI.

⭠ Return to thread

Jeremy Philip Galen

4d Great post. Curious if you've got any thoughts on the business agreements signed by developers of 'wrappers' and the underlying AI models that power their products. Bound by exclusion clauses for training and validation, the foundation models still might be subject to the same deletion reversal as the direct consumer cases you cite, right?

Like

Reply (1)

Lindsay Gross

2d Good question. The catch is "we won't train on your data" says nothing about whether it gets logged and kept, and the delete-reversal risk lives in retention, not training. The enterprise and API customers who stayed out of Judge Wang's May 2025 order were on zero data retention, and you can't preserve logs that were never written down in the first place. But that really only holds for that order since a no-retention promise is still just a contract the next one can reach. So I'd put the real exposure on retention, not the training clause.

Like

Reply