our crisis response protocol

ventr is not a crisis service, a therapy service, or a substitute for emergency support.

every message sent to june is checked, automatically and immediately, for language suggesting someone may be in crisis or at risk of suicide or self-harm. the check has two layers, and both run independently of and in addition to the AI that writes june's replies:

• a fixed detection floor: a list of crisis phrases, maintained by our team, that is matched against every message the same way every time. it does not depend on any AI model and keeps working even if the AI is unavailable.
• a separate safety classifier that reads each message for crisis signals the fixed list might miss.

either layer on its own is enough to trigger a referral.

if that check is triggered, june's usual response is replaced with a clear message pointing to real help, along with current crisis resources. this never requires the user to confirm they are safe before continuing, and the resources are never hidden behind an extra step.

resources shown may include:

• 911, if you are in immediate danger right now
• 988 Suicide and Crisis Lifeline, call or text 988 (US and Canada)
• Crisis Text Line, text 741741 (adults)
• ShelterSafe.ca, to find a shelter and its 24-hour crisis line (Canada)
• VictimLinkBC, call or text 1-800-563-0808 (BC and Yukon)
• poison centre, 1-844-764-7669 in Canada outside Québec and Nunavut, 1-800-463-5060 in Québec, 1-800-222-1222 in the United States
• National Overdose Response Service, NORS, call or text 1-888-688-6677 (Canada, 24/7)
• Health Canada's substance-use help page, for your province's 24/7 line, text, or chat
• findahelpline.com, for resources outside North America

the full current resource list is always available at /safety-resources.

when we report how often crisis referrals were shown, including the annual report California law requires, those reports will contain counts only. they will never include names, email addresses, account identifiers, or anything a person said.

june is an AI companion. she is not a licensed medical or mental health professional and cannot provide diagnosis or treatment.