MorseSign inOpen Morse

Private AI for therapy-like conversations: who can read what you tell a chatbot

The Morse teamPublished 9 min read

The most private AI for therapy-like conversations is one that keeps no readable copy of what you say, because no general chatbot falls under the rules that keep a therapist's sessions confidential. The US Department of Health and Human Services (HHS) says HIPAA applies to covered entities, such as psychologists and clinics, and their business associates (HHS), so stress, a breakup, or anxious thoughts typed into ChatGPT, Claude, or Gemini are governed by that company's own policy. As of November 2026, ChatGPT trains on Free, Plus, and Pro chats unless you turn it off, Claude trains on chats when its model improvement setting is on or a chat is flagged for safety review, and Gemini trains on future chats while Gemini History is on. Wysa, an app built for this kind of talk, sends chats to models from OpenAI, Anthropic, and Google under zero retention, and can keep your history for up to 10 years.

We read the HHS page on who HIPAA covers, the privacy and help pages of OpenAI, Anthropic, Google, and Wysa, the FAQs of Woebot and Youper, the American Psychological Association's health advisory on AI chatbots, and the Illinois, Nevada, and Utah laws on the legislatures' own sites, all on November 20, 2026. The three big chatbots earn some credit: each offers a chat that stays out of training, and each says in plain words when people can read your chats.

An AI chat isn't therapy, and this page isn't medical advice. The APA says chatbots "should not be used as a replacement for a qualified mental health care provider" (APA). If you're in crisis in the US, call or text 988, the Suicide & Crisis Lifeline, which is free, confidential, and open around the clock (988 Lifeline).

Morse is a private AI chat app with GPT, Claude, Gemini, DeepSeek, and uncensored models, encrypted so even Morse can't read your chats. It's a general chat app, not a therapy service, and it's built so talking something through leaves no readable copy on a server: your history is stored on your device, encrypted, Morse's servers keep no log of your messages, and no AI provider trains on what you send. In an E2EE chat, only a hardware enclave your browser has verified can decrypt the message.

Does HIPAA cover what you tell an AI chatbot?

No, not when you use a consumer app on your own. HHS says "the HIPAA Rules apply to covered entities and business associates," and that an organization that is neither "does not have to comply with the HIPAA Rules" (HHS). Covered entities are health care providers, health plans, and health care clearinghouses. A chatbot you downloaded is none of them, however personal the chat gets.

Professional confidentiality stays with the professional. Illinois' 2025 law keeps "all communications between an individual seeking therapy or psychotherapy services and a licensed professional" confidential (Illinois General Assembly), and a chatbot is not a licensed professional. The APA's advisory notes that licensed providers "are bound by a professional code of ethics" and answer to state licensing boards, while with a chatbot, "Users' disclosures are recorded, becoming susceptible to threats such as privacy breaches and digital profiling." For the medical side of the same gap, see private AI for health questions.

What do ChatGPT, Claude, and Gemini keep and train on?

All three keep your chats on their servers, tied to your account, and all three can train on them unless you change a setting or use a special chat. From each company's own pages, read November 20, 2026:

ChatGPTClaudeGemini
Trains on your chatsYes by default on Free, Plus, and ProIf model improvement is on, or a chat is flagged for safety reviewFuture chats, while Gemini History is on
Setting to stop itSettings, Data controls, "Improve the model for everyone"Settings, Privacy, "Help Improve our AI models"Turn off Gemini History
Chat that skips trainingTemporary chat, kept up to 30 days for safetyIncognito chatTemporary chat, kept around two months
When you delete a chatDeleted within 30 days, unless kept for security or legal reasonsDeleted from back-end storage within 30 daysReviewed chats stay for up to three years
Longest stated retentionSaved chats stay until you delete themUp to 5 years, de-identified, with training onAuto-delete defaults to 18 months, and can be set to indefinite
People who may read chatsAuthorized staff and service providers, for abuse, support, legal matters, or trainingNot described on these pages, beyond flagged chats going to its safety workHuman reviewers, including from service providers

Rating a reply undoes some of this. OpenAI says a thumbs up or down means "the entire conversation associated with that feedback may be used to train OpenAI models," even after you opt out, and Anthropic keeps a rated conversation for up to 5 years. Anthropic also says chats its systems flag as violating its Usage Policy can be kept for up to 2 years. Google's notice asks you not to "enter confidential information that you wouldn't want a reviewer to see."

Can staff read your chats, or hand them over?

Yes, in defined cases, at all three companies. OpenAI's help center says authorized personnel and service providers may access your content to investigate abuse or a security incident, to give support, "to handle legal matters," or to improve its models unless you opted out. Google says Gemini chats its human reviewers have read aren't deleted when you delete your history, and are kept for up to three years. Anthropic's privacy policy, effective September 10, 2026, allows sharing data with authorities where disclosure is needed to "comply with applicable law, regulation or legal process" or to "prevent serious harm to any person or to property."

OpenAI has written about distress directly. In August 2025, it said chats from users "planning to harm others" go to a small team that "may refer it to law enforcement" when a case involves an imminent threat, and that it was "currently not referring self-harm cases to law enforcement to respect people's privacy." Anything a company keeps can also be sought in a lawsuit, which we cover in can AI chats be subpoenaed.

What do AI therapy apps say about privacy?

Apps built for emotional support still keep your words on their own servers. Wysa's privacy policy (version 7.2.2, revised October 16, 2026) lets you start without registering, under a nickname, and says "No real people can listen to what you are talking about within the Conversation Space." The same space uses third-party language models by default, and Wysa's provider list names OpenAI, Anthropic, and Google Gemini. Wysa says it has "enabled Zero data retention with the LLM provider" and that your messages "are not used as training data by the LLM," while "We might use some anonymous chats to help our AI learn better." Where no other limit applies, it keeps data "up to 10 years from the last time you updated it," and if you came through an employer or school, it may tell that institution about an apparent emergency. Our full read is in is Wysa private.

Two well-known apps have closed. Woebot Health's FAQ says "The Woebot app was retired June 30, 2025," that previous accounts can't be accessed, and that all account data was anonymized as of July 31, 2025. Youper's home page says it "discontinued its services on September 30, 2026." When an app shuts down, its policy decides what happens to everything you told it, so read that part before you start.

Which states regulate AI in mental health?

Illinois, Nevada, and Utah each passed a law in 2025, and they regulate the service more than your data.

  • Illinois bans AI therapy without a licensed professional. Its Wellness and Oversight for Psychological Resources Act, in force since August 1, 2025, says no one may "provide, advertise, or otherwise offer therapy or psychotherapy services, including through the use of Internet-based artificial intelligence" unless a licensed professional conducts them, with civil penalties up to $10,000 per violation (Illinois General Assembly). It exempts "self-help materials and educational resources" that don't claim to be therapy.
  • Nevada bans the claim. AB 406, in force since July 1, 2025, bars an AI provider from saying, "explicitly or implicitly," that its system can provide professional mental or behavioral health care, or is a therapist, counselor, or psychiatrist, at up to $15,000 per violation (Nevada Legislature).
  • Utah protects the data. HB 452, in force since May 7, 2025, says a mental health chatbot's supplier "may not sell to or share with any third party" a Utah user's identifiable health information or "user input," with narrow exceptions such as contractors the chatbot needs to work, who must follow HIPAA's rules, and bars using your input to pick the ads you see (Utah Legislature).

None of the three limits how long a general chatbot keeps an adult's chats or whether it trains on them. Our table of AI chatbot laws by state covers the rest.

How do you talk things through with an AI and keep less on record?

Start with what you type. The chat works as well when you write "my partner" or "my manager" instead of a name, and leave out your employer, your town, and anything that would identify the other person.

Then choose the chat. Use ChatGPT's or Gemini's temporary chat, or Claude's incognito chat, for anything you don't want trained on, knowing the first two are still kept for a while. Turn off memory if you don't want a hard week carried into later chats, and check what it already saved. Don't rate replies in a sensitive chat, keep personal conversations off a work account, and set a screen lock on the device. If you'd rather write than talk, our guide to private AI journaling covers where entries go. Therapists drafting their own notes have different rules, in can therapists use ChatGPT for notes.

How Morse keeps a hard conversation private

Every service above keeps what you said on its servers, tied to an account with your email, and a policy you didn't write decides who reads it and for how long. Morse is built so your chats don't sit on a server at all.

  • E2EE, for the conversations no one else should read. Your browser checks the secure enclave's hardware attestation, then encrypts your message on your device so only that enclave can decrypt it. Morse's servers relay text they can't read, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse.
  • Private, for everyday thinking out loud. Private is the default for a new chat. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing.
  • Anonymous, to talk with GPT, Claude, or Gemini without the provider knowing it's you. Every message goes through the Morse Enclave, open-source code in AWS Nitro hardware, which sends it on under Morse's name, so the provider never learns your name, email, account, or IP address. The provider reads the message and may keep it for a limited time to check for abuse, so leave out names you'd rather keep to yourself.

In every mode, your chat history is stored on your device, encrypted, and our servers keep no log of your messages. No AI provider trains on what you send through Morse, and Morse doesn't train AI on your content, sell your data, or show ads. Anyone can check that the enclave runs the published code with npx @askmorse/verify inference.trymorse.ai (Verify). Because your chats are stored on your device, Morse has no copy of them to give authorities. A Temporary chat is kept in memory only, and you can save an encrypted backup of the rest from Settings > Data.

Try a chat from the Morse home page without an account, or start free. E2EE chats come with Pro, at $19 a month, on the plans page.

Sources

  1. OpenAI, Data controls in ChatGPT
  2. OpenAI, How OpenAI handles data in consumer services
  3. OpenAI, Helping people when they need it most, August 26, 2025
  4. Anthropic, Is my data used for model training?
  5. Anthropic, How long do you store my data?
  6. Anthropic, How do I change my model improvement privacy settings?
  7. Anthropic, Privacy policy, effective September 10, 2026
  8. Google, Gemini Apps Privacy Hub
  9. Wysa, Privacy policy, version 7.2.2
  10. Wysa, Service providers
  11. Woebot Health, FAQs
  12. Youper, Home page

All read November 20, 2026.

Common questions

Is it safe to use ChatGPT as a therapist?

Not as private as a therapist's office. OpenAI's data controls page says Free, Plus, and Pro chats help train its models by default, and its help center says authorized staff may read content to handle legal matters or investigate abuse. The APA adds that a chatbot shouldn't replace a qualified mental health provider.

Are my AI chats confidential like therapy sessions?

No. HIPAA binds health care providers, health plans, and their contractors, not a chatbot you opened yourself, and Illinois' therapy confidentiality rule names communications with a licensed professional. What you tell an AI app is covered by its privacy policy.

What's the most private AI to ask health questions?

One that keeps no readable copy of your questions on its servers, since HIPAA rarely reaches consumer chatbots. Our guide to private AI for health questions compares ChatGPT, Claude, and Gemini on health chats.

Is there an AI chat that's safe for journaling?

Look for one that stores entries on your device or encrypts them so the company can't read them, and check where the AI step sends your words. We compare Day One, Rosebud, Reflection, Mindsera, and ChatGPT in our guide to private AI journals.

Can my therapy-style AI chats be used in court?

They can, if the company stores them, because no privilege like doctor and patient covers a chatbot. Anthropic's privacy policy, for one, allows disclosure to comply with legal process, and our guide to whether AI chats can be subpoenaed covers the cases.

Will ChatGPT call the police if I say I'm struggling?

OpenAI wrote in August 2025 that it was not referring self-harm cases to law enforcement. Conversations about planning to harm others go to a small review team, which may refer an imminent threat to the police.

What happened to Woebot?

Woebot Health's FAQ says the Woebot app was retired June 30, 2025, and that all account data was anonymized as of July 31, 2025. Youper, another mood chatbot, says it discontinued its services on September 30, 2026.

Does a temporary or incognito chat keep my venting private?

It keeps it out of training, not off the servers. OpenAI may keep a temporary chat for up to 30 days for safety, and Google keeps Gemini temporary chats with your account for around two months.

Is Morse a therapy app?

No. Morse is a private AI chat app like ChatGPT or Claude, used for anything you'd ask an assistant, and it offers no therapy or clinical care. In a crisis in the US, call or text 988.

Can Morse read what I tell it?

No. Your chat history is stored on your device, encrypted, and Morse's servers keep no log of your messages. In an E2EE chat, your message is encrypted on your device so only a verified hardware enclave can decrypt it.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next