AI privacy risks are not what most people think they are. Nobody ever promised you a private conversation with a chatbot, and five different readers can retrieve what you typed without asking you first. This video walks through all five, using nothing but published policy and public court records.Switching off model training does not make your chats private. It makes them untrained on. Your conversations are still retained, still exportable, and still producible in court. On a work account, incognito mode hides the conversation from you and not from the person who pays for your seat.Covered in this video: what the training toggle actually governs, why flagged conversations and feedback buttons override the setting you carefully switched off, how shared chats ended up in Google search results, the preservation order that captured twenty million conversations people had already deleted, and what your employer can pull from an organisation data export.Then the practical part. What to change, what to stop doing, and where to put the things that should never touch a company account.
Sources referenced in this video:
* Anthropic data retention policy
* Claude incognito chats
* Gemini activity and retention
* WIRED on Claude chats in search results
* Forbes on Anthropic transcripts in Google
* OpenAI preservation order
* Wiz research on the DeepSeek database
* Chat & Ask AI leak
* Browser extension collecting AI chats
AI Fluency for Leaders:
* Three hours, CPD certified
Read more on Substack:
* Are AI chats private? Here's who can read them
Get full access to The Humans in the Loop at www.thehumansintheloop.ai/subscribe