When deleting a chat doesn't erase it

Deleting a conversation in an AI doesn't guarantee it's erased. In 2025, in the litigation between The New York Times and OpenAI, a court ordered OpenAI to preserve all ChatGPT output logs — including conversations that users had deleted and ones opened as a “temporary chat.” In other words: a legal obligation can suspend deletion policies. The only data you can be sure won't be retained is the data you never sent.

What the court ordered

On May 13, 2025, a judge ordered OpenAI to “preserve and segregate all output log data that would otherwise be deleted,” on a going-forward basis until further order. In practice, that covered even content users had chosen not to keep — by deleting conversations or using a temporary chat. OpenAI publicly objected, calling the order a privacy overreach and a break with established norms. The retention required by that order has since been lifted — but the precedent stands: a third party can be compelled to keep what you thought you'd erased.

“Deleted” doesn't mean “erased”

On most services, “delete” removes the item from your view and starts a deferred server-side erasure. But between your click and the actual erasure, several things can intervene: backups, technical logs, and — as here — legal preservation duties (a “litigation hold”) that suspend any deletion. What looks gone from your interface can remain, segregated, on third-party servers.

Why it's structural, not a one-off

This mechanism doesn't depend on a particular provider. As soon as data is entrusted to a third-party service, it enters a perimeter of obligations beyond your reach:

• litigation-related preservation duties can freeze any deletion;
• backups and technical logs outlive an interface “delete”;
• retention rules vary by country and account type.

The fix: never send the sensitive data

If retention is out of your hands once data is transmitted, the only solid guarantee is upstream: don't transmit the sensitive data. Anything that never left your browser can't be retained, produced or exposed by a third party.

1. 1Assume every prompt can be retained, whatever your settings say.
2. 2Remove the sensitive data before sending instead of relying on a future deletion.
3. 3Keep the token↔value mapping local, to stay in control of the original.

That's ONYRI Sanitize's logic: the engine detects sensitive data and replaces it with reversible tokens before sending; detection and the token↔value mapping stay in your browser, and only anonymized text reaches the AI. Whether a log is kept or not, it doesn't contain your real information.