We haven't solved how to prevent large language models from inadvertently memorizing and later leaking sensitive training data.
open
Global / Unspecified, Global
WS00774
Even with privacy safeguards, AI models can sometimes reproduce specific pieces of sensitive information they encountered during training when prompted the right way. Fully preventing this kind of unintentional data leakage remains an unresolved technical challenge.