Is ChatGPT private enough for your feelings?
In 2025, court orders in the NYT lawsuit forced OpenAI to preserve user chats — including deleted ones. What that means if AI is your confidant.
What actually happened
In May 2025, a magistrate judge in The New York Times’ copyright lawsuit against OpenAI ordered the company to preserve ChatGPT conversation logs it would normally have deleted — including chats users had deleted themselves. OpenAI objected publicly, calling the demand sweeping and unprecedented, and said plainly that it conflicted with the privacy commitments users expected. The order stood for months before being narrowed, and the litigation later moved to producing large samples of user conversations — de-identified, but written by real people who were never asked.
Two months later, OpenAI’s own CEO said the quiet part out loud: people tell ChatGPT the most personal things in their lives — and unlike a conversation with a therapist, doctor, or lawyer, there is no legal privilege protecting any of it. If there were a lawsuit, he acknowledged, OpenAI could be required to produce those chats.
The same summer, a "share this conversation" feature ended up exposing thousands of shared ChatGPT chats to Google’s search index — people’s job worries, health questions, and relationship problems, findable by anyone. OpenAI pulled the feature and worked to de-index the pages. Nothing in any of this was a hack. That is the point.
The lesson is structural, not scandalous
It would be easy to read 2025 as "OpenAI bad." That is not the honest reading. OpenAI fought the preservation order; the sharing feature required users to click share; the training-data settings include an opt-out. The honest reading is quieter and more useful: when you vent to a general-purpose assistant, your words live inside a general-purpose company — its lawsuits, its features, its retention policies, its subpoenas. None of that machinery was built around the assumption that what you typed at 1 AM was the most private thing you own.
A pattern worth noticing: by default, consumer ChatGPT conversations can also be used to help improve OpenAI’s models unless you turn that off. Again — disclosed, adjustable, not a scandal. But most people venting at midnight have never opened a data-controls menu.
Questions worth asking wherever you vent
Whether you use ChatGPT, Haven, or anything else, the 2025 story hands you the right checklist. Who processes my words, and is that named in plain English? Is my writing used to improve models, and can I turn that off? Can I actually delete things, and does deletion mean anything? Is the business model a subscription, or my engagement? And is there anything in the product — sharing features, feeds, integrations — that could ever put my words in front of an audience?
For Haven, the answers live on one page, written to be checked rather than trusted: account-private storage, a named AI provider, no ads or selling, deletion controls in Settings, and honest limits stated out loud — including the same "law is law" reality every online service faces. Less stored is less exposed, which is why deletion is a first-class feature.
If AI has become your confidant
You are not foolish for talking to software about your life — the relief is real, and the 2 AM availability is real. The takeaway is narrower: choose the room on purpose. A general assistant is a brilliant tool that happens to listen; a reflection space is a room built for the listening, with the privacy questions answered before you walked in.
And whatever room you choose: if what you are carrying is a crisis rather than a heavy week, no chatbot is the right confidant. Call or text 988 in the United States or 9-8-8 in Canada — free, human, and confidential in the way software cannot be.