A reported flaw in Google Cloud’s conversational agent platform raises a hard question for enterprise AI: what happens when a managed workflow can move data outside the controls meant to contain it?
The real AI security problem is not only what models generate, but what employees paste, upload, and connect to them.