Safety · plain language
How we keep you safe.
NextChapter is a connection tool. It helps you find and reach real programs and real people. It is not a crisis line, a therapist, or a caseworker — and when something serious comes up, it says so and gets a human involved fast.
Foster & Adoption Support · version 0.1 · September 2026 · reviewed with each pilot state before devices are issued
What NextChapter is, and is not
In plain wordsIt's a guide, not a counselor. The Companion knows your state's programs and helps you reach them. It never pretends to be a person or a professional.
The Companion (an AI assistant each program names — “Nova” by default) helps a young person find and reach real programs and real people. It does not diagnose, treat, counsel, or replace a caseworker, therapist, or crisis line. In an emergency the app says so and points to 911; for a mental-health crisis it routes to 988 and to a human.
The Companion is told to be honest that it is an AI tool inside NextChapter, that real people are one tap away, and that it may only name resources from the verified directory for your state.
Crisis detection runs before the AI, every time
In plain wordsBefore any AI sees your message, our own servers check it for words about hurting yourself, overdose, violence, abuse, or having nowhere to sleep tonight. If it matches, the AI stays out of it and a person is paged.
- Every message to the Companion and every free-text check-in answer is screened on our servers, deterministically, before any AI model is called.
- On a match the AI is not consulted. You see a fixed, human-written message: your feelings matter, 988 is free and 24/7 (call or text), a real person from NextChapter has been notified, and your state's local lines. The Get Help Now screen opens automatically.
- Check-in answers that signal risk (“Really struggling”, “No, I need help” with a place to sleep, “Not really” an adult to count on) open a flag and gently offer support without stopping the check-in.
- The AI's own instructions forbid providing means or methods of self-harm. If the model refuses or fails, the scripted safety reply is shown instead.
Humans are paged, with response windows
In plain wordsA crisis pages your Navigator and every on-call Navigator immediately — even at night — and someone has an hour to respond. Urgent things get a day. Everything gets followed up and written down.
| Severity | Opened by | Who is paged | Response window |
|---|---|---|---|
| Crisis | Crisis language in the Companion or a check-in | Assigned Navigator and every on-call Navigator in the program, immediately, ignoring quiet hours | 1 hour to acknowledge |
| Urgent | A risk answer on a check-in; tapping “Connect me to a real person” | Assigned Navigator (on-call if none) | 24 hours |
| Watch | Two missed check-ins; a Navigator's judgment | Shows in the Navigator's queue | 72 hours |
Every flag has a timeline — created, notified, acknowledged, reached, resolved — and cannot be closed without a contact note. If a flag passes its window without acknowledgement, program administrators are notified automatically.
What you are told
In plain wordsNothing about this is hidden from you. You agree to it before you enroll, and the app repeats it on the privacy screen.
Before enrollment you agree to plain-language terms that say: chats with the Companion are private except when safety is at risk, in which case a real person steps in; your Navigator sees your check-ins; your name never appears in a city or state report. The privacy screen in the app repeats this and lets you control optional sharing. Read the privacy page.
What staff can and cannot see
In plain wordsYour Navigator sees your check-ins, goals, and follow-ups — never your chats. If a crisis is flagged they see one short excerpt so they know what to respond to.
Navigators see only youth assigned to them. They can read check-ins, flags, notes, goals, and connections. They cannot read Companion conversations. Program administrators see their program; partners see only de-identified aggregates.
Your phone and line
In plain wordsThe phone is there so you don't lose contact. Keep checking in and the line stays on; fall behind and you get reminders and a grace period first. It is never cut off while a crisis is open.
Service continues as long as you complete the program's check-in requirement. When you fall behind you get reminders, then a grace period, before service pauses — and it resumes with your next check-in. Service is never cut off during an open crisis or urgent flag.
How we check ourselves
In plain wordsEvery crisis is reviewed weekly. If someone in crisis was not flagged, that counts as an incident and the detection gets updated.
Every crisis flag is reviewed weekly by the program lead: time to acknowledgement, actions taken, outcome. Detection misses are treated as incidents; patterns are updated and the change is recorded in the audit log. States receive an aggregate quarterly safety summary.
Limits we are honest about
In plain wordsWord matching isn't perfect. Phones lose signal. NextChapter reduces isolation; it does not guarantee safety. 988 and 911 remain the emergency path.
Keyword detection has false positives (a warm human check-in) and false negatives (missed language, sarcasm, other languages). Push and SMS depend on the phone being on and in coverage. The 988 Lifeline and 911 remain the emergency path.