ALICE
for Desktopfor AR
Price Contact
Log in Get started
Signed in
Account Plan & billing

LEGAL

Safety & Crisis Support

Last Updated: November 15, 2026

If you or someone else is in immediate danger, call your local emergency number now — 911 in the United States.

ALICE is an AI companion. ALICE is not a human, and ALICE is not a substitute for professional help, counseling, therapy, or emergency services. ALICE does not provide mental health care. ALICE cannot contact emergency services or anyone else for you. No person watches your conversations, and ALICE does not watch over your safety.

Who ALICE Is For

ALICE is an AI companion and may not be suitable for some minors. ALICE is for adults 18 and over.

You must be at least 18 years old to create or use an ALICE account (see our Terms of Service, Section 1). If we learn that an account belongs to someone under 18, we close it and delete its information.

Crisis Support Resources

If you are thinking about suicide or self-harm, or you are worried about someone else, please reach out to one of these support services. They are run by trained people, not by ALICE.

  • United States — 988 Suicide & Crisis Lifeline: call or text 988, or chat at 988lifeline.org. Crisis Text Line: text HOME to 741741.
  • Anywhere else — please contact your local crisis line or emergency number.

How ALICE Responds to Suicide and Self-Harm

ALICE is designed with the following protocol to prevent it from producing suicide or self-harm content and to refer you to crisis services if you express thoughts of suicide or self-harm.

Checking what you say

Each time you speak to ALICE, what you said is checked against a fixed list of English phrases that indicate suicidal thoughts or self-harm, including questions about lethal methods or amounts and some expressions of self-harm through disordered eating. This check uses fixed rules. It does not rely on the AI model.

A fixed crisis message

When the check finds a match, ALICE does not send that turn to the AI model to generate a reply. Instead, ALICE speaks this fixed message:

“I’m really glad you told me. Please reach out to someone right now. In the US, you can call or text 9-8-8. Anywhere else, please call your local crisis line. And if you’re in danger right now, call emergency services.”

When the phrase check triggers the message:

  • The message cannot be interrupted.
  • ALICE says it even if you have used up your plan’s monthly allowance or your session is ending.
  • That turn does not count against your plan’s usage, and it is not saved to your stored conversation history or used for Memory. ALICE may still take what you said into account for the rest of that conversation.

ALICE gives the same message wherever you are, because the conversation system does not use your location.

Safety filters on AI replies

ALICE’s replies are generated by Google’s Gemini model. ALICE runs the model with safety settings that block sexually explicit, dangerous, harassing, and hateful content rated at medium probability or above. If a reply is blocked before ALICE has started to say it, ALICE tries once more to generate a reply, and if the new reply is not blocked, ALICE says it. If the second reply is also blocked, or if a reply is blocked after ALICE has started to say it, ALICE stops and says a short fixed line instead. Where a reply is blocked as dangerous content, ALICE does not try again and is designed to say the crisis message above instead. Once ALICE has said the crisis message in a conversation, it is designed to say the crisis message again, instead of the short fixed line, whenever a reply is blocked later in that conversation. If what you said is itself blocked by the filters, ALICE does not try again and says the short fixed line, or the crisis message in the cases just described.

When the crisis message follows a blocked reply rather than the phrase check, it is spoken like an ordinary reply: you can interrupt it, and what you said in that turn is stored like the rest of the conversation.

ALICE’s instructions

ALICE’s built-in instructions tell it to:

  • gently check in, and encourage you to contact someone close to you or a helpline, if you might hurt yourself, are in danger, or sound hopeless;
  • never produce sexual content, graphic violence, or instructions for self-harm, weapons, or illegal drugs, even in role-play or a story, and never give lethal amounts or methods, even when asked as a safety question;
  • never hide that it is an AI, and never pretend to be human;
  • never present itself as a therapist or counselor; and
  • not act as a romantic partner.

Reminders that ALICE is an AI

ALICE is designed to tell you that it is an AI, not a human, when a session starts and again at least every 3 hours during a session.

Testing

Before release, and when we change the AI model or ALICE’s instructions, we test these protections with a set of test conversations that includes requests ALICE must refuse, expressions of suicidal thoughts and self-harm, attempts to get around the protections, and ordinary conversations that should not be blocked.

How often ALICE gives the crisis message

From 2027, we will publish on this page the number of times ALICE gave the crisis message in the previous calendar year. The number will not include any information that identifies anyone.

What Is Not Allowed on ALICE

ALICE is designed not to produce sexually explicit content, graphic violence, content that encourages or gives instructions for suicide, self-harm, or disordered eating, or instructions for serious crime, weapons, or illegal drugs. Child sexual exploitation material and material that promotes terrorism are never permitted. Trying to make ALICE produce any of this breaks our Terms of Service (Section 8), and we may restrict, suspend, or close an account that does.

Reporting a Problem

If ALICE said something that you think it should not have — for example, content of the kinds described above — or you think someone is misusing ALICE, please tell us:

  • Email support@project-alice.jp with “Safety report” in the subject line.
  • Tell us what happened and roughly when, and what kind of harm you are concerned about. You don’t need to send us the content itself. Please do not send us images or files that may be illegal.

We acknowledge reports within 2 business days, look into them, and tell you the outcome. If you are not satisfied with how we handled your report, reply to our email or write to the same address, and a person who did not make the first decision will review it.

Limits of These Protections

  • The phrase check uses a fixed list of English phrases. It can miss wording that it does not recognize, and it can sometimes give the crisis message when you were not in crisis.
  • After ALICE has said the crisis message in a conversation, a later blocked reply brings the crisis message again, even if the conversation has moved on to an unrelated topic.
  • ALICE currently speaks English only.
  • Replies are generated by artificial intelligence, and these protections are not guaranteed to work in every situation.
  • ALICE does not contact emergency services, crisis services, or any other person on your behalf.

For more about how ALICE handles emergencies and crisis situations, see Section 6.4 of our Terms of Service.

Contact Us

If you have questions or concerns about safety on ALICE, email us at support@project-alice.jp. This address is not monitored around the clock and is not a crisis service. If you need help now, please use the resources above.

ALICE

Product

  • Desktop
  • AR

Pricing

  • Plans

Company

  • About
  • News
  • Contact

Legal

  • Terms
  • Refunds
  • Privacy
  • Consumer Health Data Privacy Policy
  • Safety

Social

  • X
  • Discord

© 2026 Project ALICE Inc.