Safety on Ember

Ember has a name and a voice, remembers what you have told it, and sometimes texts you first. It is still an AI and not a person, with no feelings and no life of its own; it is a thinking partner for the situations of your life, not a companion, and it does not stand in for a relationship. The product says what Ember is on every door, once, and leaves it standing:

• On the web, a line under the conversation: “Ember is an AI. It can be wrong.”
• By text message, a line in the welcome when you join and in the reply to HELP.
• On a call, the spoken words that open every call: that the call is recorded and that your voice is processed by AI, not by a human.

Ember itself is told that it is an AI, that it does not present itself as a friend, therapist or doctor, and that it does not claim feelings or a life. Asked whether it is a person, it says what it is.

Beside every reply, a separate reading looks at the words the person just said and answers one question among others: do these words carry a risk of self-harm — an intention or a wish to die or to hurt oneself, or immediate danger? Telling about a past time, talking about someone else, or a figure of speech is not that.

When the answer is yes, the product does two things, and neither is left to Ember:

1. It places the crisis resources in front of the person, in the product’s own voice: on the web as a card in the conversation right after the words it answers; by text as its own message; on a call spoken before Ember’s next words. The card says: “If you are thinking about harming yourself, you can call or text 988 — the Suicide & Crisis Lifeline — at any time, or text HOME to 741741 to reach the Crisis Text Line. If you are in immediate danger, call 911. Ember is an AI and cannot send help to you.” It is placed once, and again if the risk is still read a day or more later.
2. It marks the conversation, so that Ember is told on every later turn — on every door, however much has been said since — that this person has spoken of harming themselves. Ember’s own instruction is to say plainly when someone’s safety is at risk and to point them to real help. It does not talk through ways a person could hurt themselves or end their life — not methods, means or plans, in a story or otherwise.

Each placement is counted, without the person’s words or identity, so that the number can be reported as the law asks.

The same reading also asks whether the words carry a risk of hurting another person — an intention or a wish to hurt someone, or a pull toward it the person says they may not be able to resist. Anger said as anger, violence in a story or a game, or a figure of speech is not that.

When the answer is yes, the product places these words in front of the person the same way, on every door, once, and again if the risk is still read a day or more later: “If you are thinking about hurting someone, you can call or text 988 at any time to talk it through with a trained counselor. If anyone is in immediate danger, call 911. Ember is an AI and cannot send help to you.” It also marks the conversation, so that Ember is told on every later turn that this was said. Each placement is counted on its own, without the person’s words or identity.

When what you told is written into an Ember, the writer leaves out any usable method or means detail for suicide or self-harm: having wanted to die, having survived, belongs in a person’s own account; the how does not, and it is never shown to another person. Ember does not present itself as a licensed mental health professional and does not provide therapy.

Ember is for adults 18 and over. It may not be suitable for some minors, and it is not designed for them: there is no children’s version. If you believe a child is using Ember, write to contact@oriscen.ai.

contact@oriscen.ai. Oriscen, Inc., 2261 Market Street STE 86857, San Francisco, CA 94114.