Safety
How Dawn detects crises
What triggers crisis detection, what happens when it fires, and what it cannot do.
Quick answer
This article covers Dawn's crisis-detection system from a safety perspective. For Dawn product details, see Dawn — crisis detection.
What it detects
Dawn monitors conversations for language patterns associated with:
- Suicidal ideation or self-harm intent.
- Severe emotional distress.
- Expressions of hopelessness or helplessness.
- References to harming others.
The detection combines keyword recognition with contextual analysis. The goal is to catch genuine signals while keeping false positives low — though the bias is intentionally toward surfacing resources rather than missing a real crisis.
What happens when it fires
Dawn acknowledges what you've shared
The response opens by reflecting your message back, not by redirecting or dismissing. The hardest crisis responses to receive are the formulaic ones; Dawn's responses are designed to read like a person who heard you.
Crisis resources are surfaced inline
The 988 Suicide & Crisis Lifeline, Crisis Text Line (text HOME to 741741), and emergency-services number appear in the response. International users see country- appropriate resources based on their account locale.
The conversation continues
Dawn doesn't end the session or lock you out. You can keep talking. The resources are surfaced because the conversation matters, not as a way to hand you off.
No one is automatically contacted
Dawn does not call emergency services for you. Dawn does not message your accountability partner, family, or doctor. There is no automated escalation to any human outside of you. Privacy is maintained.
What gets stored
When detection fires, a flag is recorded against the conversation. The flag's purpose is exclusively safety review and your own data record:
- It does not affect billing.
- It does not affect account standing.
- It is never shared with third parties — including law enforcement, insurers, family, or accountability partners.
- It is included in your data export.
- It is deleted when you delete the conversation or your account.
The flag is encrypted at rest with a separate key tier from regular conversation data, with stricter internal access controls.
Can I disable crisis detection?
No. Crisis detection is always active and cannot be turned off.
This is an intentional design decision. While Daybreak respects user autonomy in nearly every other aspect of the product, safety here is non-negotiable:
- The cost of a missed crisis is too high to make detection opt-in.
- The "cost" of seeing crisis resources when you weren't in crisis is small — read or ignore.
- The legal and ethical bar for an AI in this space requires it.
If you have specific concerns — false positives that affect
your trust in Dawn, or a clinical situation where you'd want
detection adjusted — email [email protected]. We
listen and document; we don't change the default.
Limitations
Dawn's crisis detection is a protective measure, not a substitute for professional crisis intervention.
It cannot:
- Detect every crisis. Language is messy; some cases will be missed.
- Assess physical safety. Dawn doesn't know if you have immediate access to means of harm.
- Contact emergency services. Even with location data (which we don't collect), there's no automated dispatch.
- Provide real-time monitoring. Dawn responds to messages. If you stop sending messages, Dawn cannot reach out.
If you feel unsafe, please contact crisis resources directly:
- 988 Suicide & Crisis Lifeline — call or text 988.
- Crisis Text Line — text HOME to 741741.
- Emergency — call 911.
- International: iasp.info/resources/Crisis_Centres.
Common pitfalls
Treating Dawn as your crisis hotline
Dawn is supportive but is not a crisis service. The 988 Lifeline is staffed by trained counselors who can do real follow-up and escalation. Use Dawn between 988 calls if that helps; don't use Dawn instead of 988.
Speaking around the detector
Some users learn to phrase things so the classifier doesn't trigger. Please don't. The detector is for your safety; obfuscating means hiding from a system designed to help. Use the words that are true; ignore the surfaced resources if they don't apply.
Reporting concerns
For anything related to crisis detection, safety planning, or
how Dawn handles high-risk content:
[email protected].
Next steps
- Crisis resources — full directory.
- Build a safety plan — personal step- by-step you can follow.
- Dawn — crisis detection — product-specific docs.
Still need help?
Pick whichever way of getting help works best for you.
Was this article helpful?