ChatGPT can lock chats with a passcode, without extra encryption
OpenAI is rolling out locked chats for Free, Go, Plus and Pro. You place a conversation behind a four-digit passcode or Face ID, after which it disappears from your regular history and search results. According to OpenAI the feature is not end-to-end encrypted and does not change retention periods or training settings; if you reset the passcode, all locked chats are permanently deleted.
What does this mean? A locked chat only protects against people looking at your screen, not against processing by OpenAI: client or patient data therefore still does not belong in it.
Study: AI predicts relapse in depression but with limited accuracy
A German study in JAMA Psychiatry followed 564 people with current or past depression, of whom 270 relapsed. Adding brain scans, genetics, blood values and gut flora did not improve the prediction; the best model mainly used clinical features and reached a maximum of 64.43 percent accuracy. The researchers consider that too low to base individual treatment decisions on.
What does this mean? If a supplier offers you an AI prediction of relapse risk, ask for the validated accuracy: more data does not automatically mean a better prediction.
