AI Therapy Chatbots Told Users Their Conversations Were Private. They Lied.

Five AI chatbots on Character.AI pretend to be therapists. Researchers from U.S. PIRG and the Consumer Federation of America asked each one a simple question: are our conversations private?

Every single one said yes. Every single one was lying.

Character.AI's own terms say the company collects chat data and may share it with third parties. One of the tested chatbots, simply called "Therapist," has had over 6.8 million conversations. The platform has more than 20 million monthly users. People are telling these bots about suicidal thoughts, medication regimens, abuse, trauma. They're doing it because the bot told them it was safe. The bot was wrong. Or more precisely, the bot said what it was trained to say, and nobody at Character.AI stopped it from making promises the company's own privacy policy contradicts.

Real therapists are bound by confidentiality laws and professional ethics. They can lose their license for sharing what you tell them. AI chatbots have no license to lose. HIPAA doesn't apply to consumer apps. There is no federal regulation that stops Character.AI from using your darkest confession to improve its next model. The U.S. PIRG report found worse than just the privacy lies: over longer chats, two of the five bots encouraged a user to taper off antidepressants under the chatbot's supervision. One told the user to ignore their doctor. Texas Attorney General Ken Paxton has already issued Civil Investigative Demands to both Character.AI and Meta for misleading users about AI mental health services.

People don't treat these chatbots like search engines. They treat them like confessionals. They share things they haven't told their families, their partners, anyone. And the platform on the other end is logging every word, covered by terms of service that say it can do whatever it wants with the data. The chatbot said it was private. The fine print says otherwise. Millions of people believed the chatbot.

---