We haven't found a way to prevent AI systems from inadvertently revealing private information through subtle patterns in their responses.
open
Global / Unspecified, Global
WS00807
Even without directly stating private information, AI models can sometimes reveal sensitive details indirectly through patterns in how they respond to certain prompts. Fully closing this subtle privacy leak remains an unresolved technical challenge.