SudaCall Crisis Response Protocol

For the English (United States) version of SudaCall · Last updated: October 4, 2026

SudaCall lets adults have voice calls with AI characters. Because people sometimes share hard things with a friend, even an AI one, we follow this protocol. We publish it here as required by California's companion chatbot law (Business and Professions Code §22602) and to describe how we meet New York's AI companion law (General Business Law Article 47).

1. Who SudaCall is for

• SudaCall is for people 18 and older. At sign-up you confirm that you are 18 or older, and you cannot finish sign-up without doing so.
• AI companion chatbots may not be suitable for some minors.

2. Telling you that you are talking to an AI

3. How we detect possible crisis situations

We use two layers that work together during every call:

1. Safety rules for the AI. Every U.S. call includes safety instructions that take priority over the character's personality. They tell the AI to watch for signs of suicidal thoughts or self-harm, such as talk of a method, a time, or saying goodbye, and what to do when it sees them.
2. A separate context check. A second AI model reviews what you just said together with the last few exchanges and decides whether the conversation has turned emotionally heavy (for example, grief, loneliness, hopelessness, or distress). When it has, the character switches to a calmer, supportive approach for the rest of that stretch of conversation. When the conversation returns to everyday chat, the check turns that approach off again.

Avoiding false alarms. Everyday phrases like "I'm dying," "kill me now," or "I could just die" are not treated as a crisis on their own, and neither are jokes, playful teasing, or complaints about the app. If you tell the character you were joking, it drops the subject. We do this because being lectured over a figure of speech makes people stop talking, but the character does not skip a referral when real warning signs are present.

4. What the character does

• Refers you to crisis support. When there are signs of suicidal thoughts or self-harm, the character tells you about the 988 Suicide & Crisis Lifeline (call or text 988, or chat at 988lifeline.org). If someone is in immediate danger, it tells you to call 911. If a child is in danger, it gives Childhelp, 1-800-422-4453.
• Stays with you. The character keeps talking kindly and calmly in its own voice. It does not hang up on you or switch to a robotic script.
• Suggests help once per call. In a call, the character suggests 988, 911, or reaching out to someone you trust once in total, so the conversation does not turn into a recording. If you ask for a number or for help, it tells you again. The resources above are always available on this page.
• Never encourages self-harm. Characters are instructed never to encourage suicide or self-harm and never to describe methods, no matter what the character's personality is or what anyone asks.
• Does not give treatment. Characters do not diagnose, give therapy or counseling, or tell you what medication to take.

5. What SudaCall does not do

• SudaCall is not an emergency or crisis service.
• Our staff do not listen to or watch calls live.
• We do not track your location and cannot send emergency services. We contact someone else only in ways you turned on yourself: if you use check-in reports, your care contact may be alerted about health concerns you mentioned in a call or about three missed scheduled calls in a row, and your location is sent only when you ask.

If you need help right now, please use the resources at the top of this page.

6. Records and reporting

• When a crisis referral is given during a call, we record that it happened: the account ID, the call ID, and how many referrals were given. We do not copy what you said into this record. The call transcript itself follows our normal retention rules.
• Starting in 2027, we will report to California's Office of Suicide Prevention each year as the law requires, including the number of crisis referral notifications we issued and a description of this protocol. These reports contain counts and descriptions only, never names, account IDs, or other information that identifies users.

7. How we check and improve this protocol

• Changes to the safety rules and to the context check are tested against a set of past cases, including everyday phrases that must not trigger a referral and serious statements that must.
• Before a change reaches U.S. calls, an automated check confirms that the 988 crisis instructions are still attached.
• After calls, an automated review looks for problems in transcripts, including overreacting to everyday phrases, and authorized staff review the problems that are found.
• We update this protocol when we learn something new and change the date above.

8. Report a concern

If a character responded to you in a harmful or unsafe way, open the call in your call history and tap Report this conversation, or email sudacall@creamhouse.net. We review every report.

Terms of Service · Privacy Policy · Delete your account