Is it safe to tell ChatGPT personal things? What happens to each kind

The Morse teamPublished October 27, 20269 min read

Whether it's safe to tell ChatGPT personal things depends on what kind of thing it is. As of October 2026, a health question, a money worry, or a secret about a friend is handled like any other chat, but OpenAI runs separate systems for two kinds of personal talk. Signs that you might hurt yourself can bring crisis resources, a human review, and a short alert to a person chosen in advance, though OpenAI says it doesn't send adults' self-harm cases to the police. Plans to hurt someone else can bring a review, a ban, and a call to the police.

We read OpenAI's safety posts from August 2025 to May 2026, its help pages on consumer data, memory, data controls, Health, and trusted contacts, its privacy policy, and its policy for government requests, all on October 27, 2026. To OpenAI's credit, its pages spell out what sends a chat to a human reviewer and what an alert does and doesn't contain.

This page is about privacy, not medical advice. If you're thinking about suicide or self-harm, you can call or text 988 in the US, any time, to reach the Suicide and Crisis Lifeline.

What can happen to each kind of personal thing you tell ChatGPT?

Most of it stays on OpenAI's servers until you delete it, open to authorized staff and service providers for reasons we cover in is ChatGPT private, and used for training if that setting is on. Rating a reply undoes your training opt-out for that chat, since OpenAI says a thumbs up or down means "the entire conversation associated with that feedback may be used to train OpenAI models" (OpenAI). Here is each kind, from OpenAI's pages as read on October 27, 2026.

Is it safe to tell ChatGPT about suicidal thoughts or self-harm?

It's more private than you might fear, but it isn't unwatched. In August 2025, OpenAI wrote: "We are currently not referring self-harm cases to law enforcement to respect people's privacy given the uniquely private nature of ChatGPT interactions" (OpenAI). None of the 2026 safety posts we read says that has changed.

By OpenAI's account, two things do happen:

• ChatGPT changes how it answers. It is trained not to give self-harm instructions and to point people to crisis lines: the US line above, Samaritans in the UK, and findahelpline.com elsewhere (same post). Since September 2025, OpenAI has routed some sensitive conversations, such as signs of "acute distress," to a reasoning model "regardless of which model a person first selected" (OpenAI).
• You're warned before a person looks, if you've named a trusted contact. OpenAI's help page says that when a conversation is flagged for review, ChatGPT tells you it may notify your contact, and "If you do not receive this notification, the conversation will not be sent for human review and the trusted contact will not be notified" (OpenAI). How trusted contacts work, and the "safety summaries" from earlier chats that OpenAI says are "kept only for a limited time," are covered in is ChatGPT private.

You wouldn't be unusual. In October 2025, OpenAI estimated that about 0.15% of users active in a given week have conversations with "explicit indicators of potential suicidal planning or intent" (OpenAI).

Does ChatGPT report you to the police?

For threats to other people, it can, and OpenAI's wording has firmed up. In August 2025, OpenAI said conversations from users "planning to harm others" go to "a small team trained on our usage policies," and that if reviewers find "an imminent threat of serious physical harm to others, we may refer it to law enforcement" (OpenAI). OpenAI's April 28, 2026 post on community safety drops the "may": "When conversations indicate an imminent and credible risk of harm to others, we notify law enforcement." It calls its criteria "flexible," because a user "may not explicitly discuss the target, means, and timing of planned violence" (OpenAI).

Venting is a different thing from planning, and OpenAI says it trains ChatGPT to tell them apart. The same post says people "express fear or anger, or talk about violence in ways that are fictional, historical, political, personal, or potentially dangerous," and that a flagged conversation is "assessed in context by trained personnel" who can dismiss it as low risk. Where it isn't dismissed, a ban can reach "other accounts of the same user."

Police can also ask OpenAI for your data, which we cover in AI transparency reports.

Will ChatGPT remember what I told it?

Probably, if memory is on, even if you never asked it to. Since a June 2026 update, memory is built by a background process OpenAI calls dreaming, which "makes it easier for memory to include context that occurs naturally in conversation, without relying on explicit requests to remember something" (OpenAI). What deleting a chat leaves in memory, and how to clear each source, is in is ChatGPT private.

Is AI therapy private?

Not in the legal sense, and OpenAI's chief executive wants that changed. In a September 2025 post, Sam Altman wrote that because talks with a doctor or a lawyer are privileged, "the same level of protection needs to apply to conversations with AI," and that OpenAI is "advocating for this with policymakers" (OpenAI).

The same post lists what would stay open even under that privilege: "automated systems will monitor for potential serious misuse," and "threats to someone's life, plans to harm others, or societal-scale harm" may be "escalated for human review." So the privilege OpenAI is asking for would still leave the self-harm and harm-to-others checks above in place. Altman has also said "we haven't figured that out yet," as TechCrunch reported, and how courts have handled AI chats so far is in can AI chats be subpoenaed.

Is it OK to tell ChatGPT about other people?

You can, but the other person gets no say. OpenAI's privacy policy collects the "Personal Data that you provide in the input," and that includes whatever you write about your sister, your manager, or your ex (OpenAI). It stays under your account, with your training setting and your memory.

OpenAI's May 2026 privacy post offers two protections, both partial. Before training, an internal version of OpenAI Privacy Filter runs "on user conversations if they have 'Improve the model for everyone' enabled," and it "identifies and masks personal information in text." ChatGPT is also "designed to reject requests for private or sensitive information about individuals" (OpenAI). Neither stops staff from reading what you wrote. Naming a trusted contact sends their name, email address, and any phone number you add to OpenAI too (OpenAI).

Leaving the name out costs you nothing. ChatGPT answers "my friend told me she's pregnant" as well as it answers the same sentence with her full name.

Can parents see their teen's ChatGPT chats?

No, but they can be alerted. OpenAI's April 2026 post says "Parents don't have access to their teen's conversations," but where "our system and trained human reviewers detect possible signs of acute distress," a linked parent may be told "only with the information needed to support their teen's safety," by email, text, or push alert (OpenAI).

The alerts have widened. A July 13, 2026 update added notices when a linked teen's account "is banned for violent activity," and says they aren't meant for "fictional writing, gaming, news discussion, political discussion, general anger, or abstract questions" (OpenAI). In August 2026, OpenAI added notifications "related to eating disorders" (OpenAI). For a teen with suicidal thoughts, Altman wrote in September 2025 that OpenAI "will attempt to contact the users' parents and if unable, will contact the authorities in case of imminent harm" (OpenAI).

How Morse keeps your personal chats to yourself

Every row in the table starts on OpenAI's servers, where ChatGPT keeps your chats under an account with your name and email, and where staff, reviewers, and a court can reach them. Morse is built so there's no such archive of yours.

• No server copy to read or remember. Your chat history is stored in your browser, encrypted, and locked when you sign out. Morse's servers don't keep your messages or log what you write.
• No training. Morse doesn't train AI on your chats, sell your data, or show ads, so there's no training switch to find.
• Very little to hand over. Because we don't store chats, we usually have very little to give authorities.
• Crisis lines in character chats. Morse shows crisis lines when a message in a character chat reads like suicidal thinking, and that check runs on your device.

Then pick the mode for what you're about to say:

• E2EE, for the rows in the table you'd tell no one. Before a word leaves your device, your browser verifies the enclave's hardware attestation and encrypts the message for that enclave alone. Morse's servers pass along ciphertext they can't read, and the reply is decrypted only in your browser. E2EE uses open-weight models and depends on the enclave hardware and attestation working as designed, as we explain in how end-to-end encrypted AI chat works.
• Private, the default, for a health worry or a money question you'd rather no host kept. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, answer under Morse's zero data retention agreements with its inference, search, and other providers.
• Anonymous, for GPT on everyday questions. Our proxy strips your account, email, and IP address and sends the message under Morse's own account, so OpenAI reads the words without knowing it's you. It may keep them for a limited time to check for abuse, under its own policy, so names, yours or anyone else's, are worth leaving out. More in how to use ChatGPT anonymously.

Clearing your browser, losing the device, or deleting your account deletes your chats, and Morse can't recover them, so copy anything you'll want later into your own notes.

So what should you tell ChatGPT?

• Health or money: Ask without names, record numbers, or account numbers, and turn off "Improve the model for everyone" first. On Morse, start the chat in E2EE.
• Someone else's secret: Leave out who it is. On Morse, your chat history stays in your browser, not on Morse's servers.
• Hard feelings: ChatGPT is trained to answer them with care, and by OpenAI's account an adult's self-harm talk isn't sent to police. If you want it somewhere no one else can read it, write it in E2EE on Morse, and reach a crisis line if you're in danger.
• Anger: Venting is fine, but anything that reads like a plan can reach a reviewer.

Try a chat from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.