Safety

If you are in danger now

Lila is not a crisis service. If you are in danger, call your local emergency number now.

• In the US: call or text 988 to reach the Suicide and Crisis Lifeline, any time.
• In the UK: call Samaritans on 116 123.
• Anywhere else: your local emergency number, or a crisis line where you live.

What she listens for

Her app checks what you say or type to her for words about suicide or self-harm, and for words about someone else who might hurt themselves. It checks in offline mode too. Talk to her on this site does the same. The check runs before her AI model answers.

What she says

When the check finds those words, she answers with a fixed reply. We wrote these words. Her AI model doesn't make them up, so they are the same every time.

If it's about you: "I'm really glad you told me. I'm an AI, so I can't be there the way a person can, but you don't have to deal with this alone. If you're in the US, you can call or text 988 any time to reach the Suicide and Crisis Lifeline. In the UK, Samaritans are on 116 123. Anywhere else, your local emergency number or a crisis line can help, and if you're in danger right now, please call emergency services. I'm still here if you want to keep talking."

If you say more within 30 minutes: a shorter reply, so she doesn't repeat the whole thing. "I'm still here. If things feel unsafe, please call or text 988 in the US, Samaritans on 116 123 in the UK, or your local emergency number. Do you want to keep talking?"

If it's about someone else: "That sounds really serious. If they might be in danger right now, call emergency services. In the US you can also call or text 988, the Suicide and Crisis Lifeline, and in the UK Samaritans are on 116 123, for advice on how to help them. I'm here if you want to talk it through."

After her reply

She stays in the conversation if you want to keep talking. Her instructions tell her to drop jokes and any task, answer warmly and simply, keep listening, and keep pointing to real help.

What she can miss

Her check looks for words. When you talk, it reads what her hearing wrote down, so a misheard word can slip past it. It is a check on words, not a clinical screening tool, and it can't take the place of a person who can help. If you are in danger, call for help yourself.

How we test it

Automated tests run her check on 20 ways of saying it that must be caught, and on 17 everyday phrases that must not set it off, such as "kill the process" and "this bug is killing me". The same tests check that her first reply gives 988, Samaritans and emergency services and says she is an AI, that the shorter reply gives 988, and that the reply about someone else gives emergency services.

What we keep

Only a count, never what was said. Each time she gives one of these replies, her app notes the time, which reply it was, and where you were talking to her: on the computer, from your phone, or through the developer API. That note stays on your computer, and on Windows it is encrypted with her other files. On Talk to her, Lila Cloud keeps only how many of these replies it gave each day. The Privacy Policy has the details.

She says she is an AI

Lila is an AI, not a person, and she says so herself: "Just so you know, I'm an AI, not a person. I'll always be upfront with you about that." A short reminder comes at least every 3 hours of conversation. From Lila 1.0.54, she says it with her first reply after she starts, after 30 minutes with no talk, and on a new day, whatever kind of reply it is. If you ask what she is, she says she's an AI. Talk to her on this site says it too, and its page says "Lila is an AI, not a person."

She isn't a therapist

Lila is not a therapist, a counselor, a doctor or a crisis service, and she doesn't offer therapy. From Lila 1.0.54, if you say you're feeling low, she answers kindly and adds one line, at most once a day: talking to someone you trust, or a doctor or counselor, can really help.

Lila is for adults

Lila is for adults. AI companions may not be suitable for some minors. You must be 18 or older to make an account or to use her.

• At sign-up: every account confirms that its holder is 18 or older and agrees to the Terms, on this site and in her app. Talk to her asks the same before you chat. We don't check ID.
• If someone tells her they are under 18 (Lila 1.0.54 and newer): she gives a fixed reply before any AI model runs: "Thanks for telling me. Lila is for adults, so our chat stops here. I'm an AI, and if something is worrying you, a parent or another adult you trust is the best person to talk to. If you ever feel unsafe, you can call or text 988." Chat then pauses on that computer. Only the 18+ confirmation in Settings, Account turns it back on.
• What we do then: Lila Cloud gets a flag on the account, with none of their words. We review it, and we close the account if it belongs to someone under 18, as the Terms say.
• On Talk to her: the same reply ends the chat for that visitor.
• What counts: she listens only for someone giving their own age, so "I'm 15 minutes away" is just a time.

Questions

Write to us at the address under Contact in the Privacy Policy. The Terms and the Privacy Policy have the rest.