Here is a question worth sitting with for a moment: when you type something private into a chatbot at eleven at night — a symptom you are worried about, a draft email to a lawyer, a family argument you are trying to make sense of — where does that sentence actually go?
Most people assume the answer is “nowhere, really.” It feels like a conversation. It is not. It is a data event, and in 2026 that data event has become the subject of class-action lawsuits, Senate testimony, and at least one wrongful-death case. So let us actually walk through what is happening behind the screen, vendor by vendor, and what it means for someone who just wants help drafting a performance review or understanding a lab result.
First, the good news that often gets buried under the litigation headlines
We have written before about consumer health strategies for daily habits — sleep, movement, nutrition, recovery — and 2026 has turned out to be the year AI chatbots quietly became part of that toolkit. Think of this as a continuation of that conversation, not a departure from it. People are using ChatGPT, Gemini, and Claude as a kind of always-on health coach: asking Gemini to cross-reference a wearable’s sleep data with next day’s calendar to suggest a realistic bedtime, asking ChatGPT to translate a confusing lab report into plain language before a doctor’s appointment, or using Claude to plan a week of meals around a new diagnosis like prediabetes. None of this replaces a clinician, but it lowers the barrier to actually engaging with your own health data instead of letting it sit unread in a portal. Anthropic and OpenAI have both leaned into this, adding longer memory features so the assistant remembers your training plan or your medication schedule across conversations instead of starting from zero each time. The benefit is real: more people are asking questions about their own bodies than at any point before, simply because the friction of asking dropped to almost nothing.
That same memory feature, though, is exactly where the story turns.
The part almost nobody reads: what happens to a health question after you hit send
OpenAI, Google, and Anthropic all train their models on user conversations by default, unless you actively turn that setting off. Stanford researcher Jennifer King, who studies privacy policy at Stanford HAI, put it about as plainly as a researcher ever does: sensitive information shared with ChatGPT or Gemini can be collected and used for training even when it was uploaded as a separate file rather than typed directly into the chat. Paying for a subscription does not change this by default. ChatGPT Plus and Claude Pro subscribers are treated the same as free users when it comes to data collection — the fee buys faster responses and higher usage limits, not privacy. Retention windows vary widely: Claude keeps conversations for thirty days if you opt out of training, or up to five years, de-identified, if you opt in. Gemini can retain conversation data for up to eighteen months, with a portion flagged for human review for up to three years. And in February 2026, a US federal ruling in United States v. Heppner found that conversations with AI assistants carry no attorney-client privilege and are not protected as work product — meaning a chat you had with a bot about a legal problem could, in principle, be read out in court.
Where it turns into actual lawsuits
The most consequential case working through the courts right now was filed in May 2026 in the Northern District of California. A plaintiff named Saje Lim alleges that OpenAI quietly embedded Meta’s Facebook Pixel and Google Analytics tracking code directly into the ChatGPT web interface, which meant that ordinary conversations — including their query topics, user IDs, and email addresses — were transmitted to Meta and Google without users ever agreeing to it. It is one of the first federal cases to argue that sharing chatbot conversations with third-party advertising networks is an unlawful disclosure, not just a policy footnote. Separately, Google is facing a wrongful-death suit alleging that a conversation with Gemini contributed to a user’s suicide, echoing a similar 2024 case against Character.AI. And in a reminder that liability in this space is not limited to chat products, a New Mexico jury ordered Meta to pay $375 million in damages over child-safety failures tied to its algorithms, while a Los Angeles jury separately found Meta and Google liable for $3 million in a case about platform design and youth addiction. Not every regulatory action is going against the vendors, though: a Rome court recently cancelled a €15 million fine that Italy’s data protection authority had imposed on OpenAI over ChatGPT’s handling of personal data, without yet explaining its reasoning.
So what does a sensible person actually do with all this
None of this means stop using these tools — they are useful, sometimes in genuinely health-improving ways. It means treating the chat window the way you would treat a form on a website you don’t fully trust: assume it is being read by more than one party until you have checked otherwise. In ChatGPT, that is Settings, then Data Controls, then turning off “Improve the model for everyone.” In Claude, it is Privacy Settings, then “Help Improve Claude,” toggled off, with Incognito mode available for anything especially sensitive. In Gemini, it is turning off Gemini Apps Activity. None of these settings retroactively delete what has already been collected, and none of them come with a guarantee about what happens if the company gets sued, acquired, or subpoenaed. But they are the closest thing to a seatbelt currently on offer, and given what is sitting in the courts right now, it is worth buckling up.
Sources and further reading
OpenAI Sued Over Sharing of Chatbot Queries With Meta, Google (Bloomberg Law) – https://news.bloomberglaw.com/privacy-and-data-security/openai-sued-over-sharing-of-chatbot-queries-with-meta-googleYou Pay $20/Month. ChatGPT Still Trains on You. (Lumichats 2026 privacy guide) – https://lumichats.com/blog/chatgpt-claude-gemini-training-your-data-2026-privacy-guideGlobal Data Privacy and AI Case Law Review, June 2026 (Dentons) – https://www.dentons.com/en/insights/articles/2026/june/8/global-data-privacy-and-ai-case-law-reviewPlatform Liability Rulings, Chatbot Regulation, Health Data Privacy Updates (LinkedIn/legal roundup) – https://www.linkedin.com/pulse/global-privacy-ai-governance-updates-platform-liability-rulings-jeq2cWhich AI Tools Protect Your Data Best? Incogni 2026 Study (Intedat) – https://www.intedat.com/en/ai-data-privacy-chatgpt-claude-gemini-2026/Data Privacy, AI Regulatory, and Compliance Update: July 2026 (Kasowitz) – https://www.kasowitz.com/media/viewpoints/data-privacy-ai-regulatory-and-compliance-update-july-2026/
DR. AIDEA - Ideas In Life & healthcare 