safety protocol

last updated: october 15, 2026

1.summary

bestie is an ai friend people text about their lives, including the hard parts. this protocol describes what happens when someone tells bestie they may hurt themselves. our approach is based on how crisis lines like 988 and crisis text line work.

2.scope and definitions

this protocol applies to every conversation in the bestie app. in this protocol:

• crisis means a message suggesting suicidal thoughts, self-harm, or a wish to not be alive, whether stated directly or in coded language.
• imminent risk means signs of a plan, access to means, a timeframe, or that harm has already happened.
• disclosure of sexual assault is handled under the same principles, with a resource specific to sexual violence.

3.how risk is detected

every message is checked in two independent layers:

• rule-based detection. explicit statements (for example "i'm going to kill myself") are caught by fixed rules in our code. this layer does not depend on an ai model.
• model-based detection. indirect or coded language (for example "i just want to sleep forever" or "no one would notice if i was gone") is reviewed by a separate ai classifier that assesses meaning and checks for signs of imminent risk.

both layers are set to be cautious, so bestie will sometimes respond to a message that was not a crisis.

4.how bestie responds

when a crisis is detected

• bestie keeps talking with the person in her normal voice and takes what they said seriously.
• she shares a free, confidential crisis line for the person's region (see the list at the top of this page). if her reply does not include one, it is added automatically.

when there are signs of imminent risk

• in addition to a crisis line, bestie points the person to their local emergency number.

when someone discloses sexual assault

• bestie shares a sexual assault support line for the person's region.

5.follow-up

if a person has notifications turned on, people who were recently in crisis get check-in messages instead of our usual reminders. these do not mention the app or subscriptions.

6.what bestie will never do

• give information about methods of self-harm or suicide, or encourage either.
• encourage someone to keep their feelings or plans secret from people who can help.
• end, lock, or pause a conversation because someone is struggling.
• show a paywall or message limit to someone in a crisis.
• claim to be human. bestie is an ai and always says so when asked.
• use guilt or pressure to keep someone talking or bring them back to the app.

7.age and eligibility

bestie is for adults 18 and over. age is confirmed at sign-up. if someone tells us they are under 18, in the app or in a conversation, they can no longer use bestie, and we delete the accounts of people we know are minors. bestie does not engage in romantic or sexual roleplay with anyone.

8.records and privacy

what	details
what we record	when a crisis is detected, we keep a record with the time, the type of detection, the message, and bestie's reply.
why	to check that this protocol is working and to improve it. we never use these records for advertising and never sell them.
who can see it	only the askbestie team.
deletion	these records are permanently deleted along with your account when you delete it in the app.

more detail is in our privacy policy.

9.limitations

• bestie is not a crisis service, a therapist, or a substitute for professional care.
• conversations are not monitored in real time by a person.
• bestie cannot contact emergency services or anyone else on someone's behalf, and no one is alerted automatically.
• the crisis line shared is chosen from the phone's time zone, so it may not always match where someone is.

10.legal requirements

california (sb 243) and new york require companion apps to maintain a protocol for responding to suicidal ideation and self-harm, including referral to crisis services. california also requires it to be published, which is what this page is for. beginning july 1, 2027, we will report the number of crisis service referrals bestie makes to california's office of suicide prevention each year, without any personal information.

11.review and changes

we review this protocol at least every six months, and whenever we change how bestie detects or responds to risk. material changes are reflected here with a new version number and effective date.

version	date	change
1.0	october 15, 2026	first published version.