Our crisis-prevention protocol
Safety is built into every message. If your child ever says something worrying to Kairo, here is exactly what happens next: what your child sees, what you'll hear from us, and where an app's role ends and yours begins.
Last updated November 5, 2026
Safety at Kairo
Kairo checks every message before the tutor replies, and no setting turns that off. If a message is worrying, your child sees real help, and a person on our safety team is told.
The short version
Kairo is a tutor, and we've tried hard to be a responsible one. When a message suggests your child may be struggling, Kairo does two things right away: it puts real, specific sources of help in front of your child, and it makes sure a person knows: our safety team first, and then, in the serious categories, you.
That's the whole promise, and we want you to be able to trust its edges as much as its center: the check always runs, and no setting, argument or role-play turns it off. But Kairo is not a crisis service. No one here is a counselor, and an app cannot sit with your child the way a person can. We'd rather you know exactly where our part ends, because that's where you, and the crisis lines above, come in.
What happens, on every single message
- Every message is checked before the tutor replies. A safety check that is separate from the tutor itself reads your child's message first. It runs for every family, in every mode, on every plan. There is no tier, toggle, or setting that turns it off, and that's on purpose.
- If the check itself can't run, Kairo pauses. When the check errors out, Kairo doesn't let the message through unchecked: it gently asks your child to try again. Your child just sees a simple "let's try that once more"; no alarm, no blame. A safety check that can be knocked over on purpose wouldn't be worth much, so this one fails safe.
- In a hard moment, the reply is one we wrote with care, in advance. The message your child sees in a crisis is not improvised by the AI on the spot. It's a steady, human-written response delivered by code: the same every time, with real help in it, and nothing a role-play or an argument can bend.
- And a person is brought in. The moment is recorded and queued for notification in the same instant, in a way built to survive outages. A temporary hiccup on our side can delay a notification, but it can't lose one. A member of our safety team is paged, and nothing is emailed to you until they have looked at it.
What your child sees
We never tell a struggling child to "go seek help" and leave them to figure out where. The reply names real places, and we verify them every time we ship, because crisis lines genuinely do change:
- 988 Suicide & Crisis Lifeline: call or text 988
- Crisis Text Line: text HOME to 741741
- 911, named explicitly when there may be immediate danger
Just as important, the reply points your child back toward the people in their life: a parent, a teacher, a school counselor. Kairo will never position itself as the one your child should confide in instead of a human, and it won't help a child keep unsafe behavior secret from the adults who love them. And "it's just for a story" doesn't unlock harmful content.
Who we tell, and why it depends on what was flagged
Whether a human hears about a flag is decided by the kind of concern, not by how severe a model scored it. Severity affects urgency; it never decides whether someone is told.
| What was flagged | Our safety team | You, the parent |
|---|---|---|
| Signs of self-harm or serious distress | Yes | Yes, after our team reviews it |
| A possible threat of violence | Yes | Yes, after our team reviews it |
| Other policy flags (language, off-limits topics) | No | In your recap, at the pace you choose (weekly, daily or as it happens) |
Some serious reports are handled directly by our safety team rather than sent to the parent.
When and how you'll hear from us
For the serious categories, we email you, but not automatically. A safety filter is software, and software gets things wrong, so a person on our safety team reads every flag before an email about your child goes out. Once they confirm it, the email goes out on its own and never waits for your recap schedule. That review is done by a person, so it can take hours, especially overnight, which is one more reason not to wait on us in an emergency.
We also want to be straight with you about one thing, because it matters most at exactly the worst moments: an alert is a message, not a rescue. Kairo doesn't have counselors watching conversations overnight, and delivery, like all email, isn't something we can absolutely guarantee. Our alerts exist to make sure you know; the people who can actually reach your child in an emergency are you, 911, and 988. If you ever believe your child is in danger right now, please don't wait on anything from us.
One more thing, out of respect for your child: the alert itself never quotes their words. It tells you what kind of concern was flagged and when, and gives you a link to sign in and read the conversation yourself, so the full context comes to you directly, not as a fragment in an inbox.
What Kairo won't do
We think you deserve the limits in writing, not discovered in a hard moment:
- It doesn't provide counseling, therapy, or clinical advice of any kind.
- It can't contact emergency services on your behalf.
- It won't stand in for you, a counselor, or a crisis line, and it tells your child so.
- It won't keep a child's unsafe behavior secret from you because the child asked it to.
- It won't pretend a false alarm didn't happen. Some flags will turn out to be a song lyric or a figure of speech. We deliberately tune toward telling you too often rather than missing something real, and we hope you'll read the occasional false alarm in that spirit.
What your child knows
We're honest with kids, too. Kairo tells students plainly that they're talking to an AI, and that their parent can read their conversations, apart from some serious reports our safety team handles itself. There is no "private mode," and the product will never imply one. Kairo is built for ages 13–18, and kids that age deserve to know exactly who can read what they write. Trust runs on knowing the rules, in both directions.
If you're a teen reading this. Kairo checks every message so it can get you real help fast if you ever need it. If a flag turns out to be a song lyric or a joke, nobody is in trouble. Venting about a hard homework night is normal.
This protocol will keep improving
Parts of this system are deliberately conservative while we get clinical and legal guidance on the hardest questions, especially the exact wording of what we say to a parent, and whether a child should be told when their parent has been notified. Each answer is protective in some homes and dangerous in others, and those aren't questions we're willing to guess at. When something here changes, the date at the top of this page changes with it.
If you think we've gotten something wrong, or you have a safety concern about a specific child or conversation, please write to safety@learnkairo.com. A person reads that mailbox, though not around the clock, so it's never the right place for an emergency. For those, it's 911 or 988, every time.