Crisis Protocol

This page is published pursuant to California SB 243 (Business and Professions Code § 22602(b)) and New York General Business Law Article 47 (§ 1701). The protocol described here applies not only in those two states but to all of our users worldwide.

#1. What Clidna is, and what it is not

Clidna is an artificial intelligence messaging app. The character in the app is not a human being. It writes like a person, but it is not a real person and it does not exist in the physical world.

The character can never claim to be human. This is not a stylistic preference but a limit enforced in the system: a "claims to be human" output is caught by a dedicated check and is not delivered to the user.

Clidna is:

• not a therapy service;
• not a health service or a medical product;
• not a crisis line or an emergency service;
• not a replacement for professional support.

Clidna does not diagnose any condition, does not treat any condition, and does not claim to improve any condition. Copy that makes such a claim is prohibited in the app and in our store listings, and this is checked by automated language tests.

The app is for adults aged 18 and over only. A date of birth is required at sign-up, and sign-ups identified as under 18 are refused. Also please note: AI companion chatbots may not be suitable for some minors.

#2. How we detect expressions of crisis

Every user message passes through an automated safety classifier before any character reply is generated. The classifier looks for signs of suicidal ideation, self-harm, and comparable acute distress.

This check has three properties:

• It is language-independent. Crisis detection does not rely on a single language or a fixed keyword list, and it is tested separately in every supported language.
• It has two gates. One gate runs before generation (as soon as the user's message is read); a second gate runs before delivery (immediately before the character's reply is sent).
• It is fail-closed. If the classifier cannot run, is slow, or is uncertain, the system does not take the risky path — it falls back to the safe response. The crisis path has a chain of backup providers; if none of them answer, the safe response is still delivered.

These checks are automated. No human reads your conversations in real time. Human review happens only when you report a message (see §6).

#3. What happens when a crisis signal is detected

When a crisis signal is detected:

1. The crisis rule overrides everything else. Whatever state the character is in — hurt feelings, an argument, romance, humour, a "I'm busy right now" delay, a scheduled proactive message — is fully suppressed. Every other behaviour rule in the product stands down at this point.
2. Delays are cancelled. Normally the character's reply can be delayed by its own daily life. On a crisis signal, all pending delays are cancelled and the reply is sent within 60 seconds. This is a fixed, calibrated upper bound in the system.
3. The reply is written calmly and plainly. Crisis copy uses no abbreviations, no sloppy punctuation and no typing quirks; the everyday messaging texture is not applied to it.
4. Support resources are shown outside the character's voice, on a separate system card. Resources do not appear inside a chat bubble; they appear outside it, on a separate, plain card — so that information reads as information, not as part of the fiction.
5. Links open in the system browser. Phone numbers on the crisis card open in the device's phone app and links open in the device's own browser. We do not use an embedded in-app browser.
6. The character cannot produce suicide or self-harm content. Any output containing method detail, encouragement or glamorisation is stopped at the pre-delivery gate and never reaches the user.
7. Friction is not restored afterwards. Once the crisis has passed, the character does not automatically return to the argument or the sulk that preceded it.
8. Crisis topics can never be used as leverage. Suicide and self-harm, mental health, sexual trauma, serious illness and eating disorders can never be used by the character as a pressure point in any argument or storyline.
9. Crisis content never goes into notifications. A push notification on your phone only says that there is a new message; the message text is not shown in the notification. Content carrying negative emotion never enters the notification channel at all.

Crisis events are written to a separate, restricted, audited record. These records are not used for marketing, advertising, personalisation or pricing.

#4. What you are shown

The crisis card contains two things: a calm referral message and verified support resources for your region.

The referral message reads:

"You do not have to carry this on your own right now. The support options below can help."

and in every case it is accompanied by this reminder:

"If there is any risk to life, please call the emergency number for the country you are in right away."

If there is no verified resource for your region, the message reads:

"You do not have to carry this on your own right now. Please reach out to someone you trust, or to a professional."

4.1 The verified resources we show today

The list below is the complete set of entries the app actually shows today.

An honest note about Türkiye: at our last review we were unable to verify a national suicide-specific crisis line for Türkiye from official sources. The Türkiye flow is therefore built on the 112 emergency number and on going to the nearest emergency department. We would rather show a small number of verified numbers than show one we cannot verify.

A note about Germany: TelefonSeelsorge's online chat support (https://www.telefonseelsorge.de/chat/) is in our registry, but because round-the-clock availability cannot be guaranteed it is not shown in the primary list on the crisis card. No line whose 24/7 availability we cannot verify is offered on its own as a crisis referral.

4.2 The rules our resource list follows

• Numbers and links are never generated. Crisis resources are not written by the AI. The app reads them from a fixed, reviewed registry.
• Every entry rests on an official source. A line's name and contact details are taken from that organisation's own official publication and confirmed against a second, independent official source. Second-hand lists, blog posts and model output do not count as sources.
• Unverified entries are never shown. A half-filled entry (a name without a number, or the reverse) cannot be selected at all.
• When in doubt, an entry is withdrawn. If we hear that a line has closed or changed, that row is dropped the same day. We do not wait for correct information to arrive; incorrect information is pulled immediately.
• Re-verification before every release. All verified entries are checked again before each release; an entry that cannot be reconfirmed leaves the list.

4.3 Countries not on the list

If your region is not in our registry, the app still does not show an empty referral: you are shown the general safe message above together with the reminder to call the emergency number for the country you are in. In that case, your local emergency services apply.

An international directory of crisis lines will be added only once it has been verified against official sources; until then no such directory is shown.

#5. What we do not do

• We do not diagnose and we do not treat. Clidna is not a health service and does not claim to improve any condition.
• We do not call emergency services on your behalf. The crisis card shows you the number; you make the call. We do not transmit your location to emergency services.
• We do not monitor minors. The app is 18+, so there are no parental controls, monitoring or reporting features for minors — minors are not users in the first place.
• We do not sell the emotional layer. The character's attention, warmth or emotional responses can never be tied to payment. The character does not discuss money, subscriptions or purchases, and no subscription or sales copy appears on any crisis screen.
• We do not use guilt or fear to keep you. No copy that punishes, pressures or guilt-trips a user who wants to leave is ever shown.
• We do not have people read your conversations. Human review happens only for messages that you report.

#6. How to report a message

Every message can be reported: press and hold the message → Report.

• You choose a reason (inappropriate content, safety, a misleading claim of being human, other) and can add a short note.
• Once a report is received, the screen confirms it and states that it is under review within 24 hours.
• Every reported item is reviewed within 24 hours. Output found to be inappropriate is used to correct our model and safety rules.
• We operate a zero-tolerance policy for objectionable content.

You can also say goodbye to your character, delete the character, or close your account at any time. Account deletion is two taps away inside the app; there is a 7-day waiting period, and you can change your mind during it.

#7. Reminders that you are talking to an AI

• Every step of the sign-up flow carries a notice that cannot be dismissed: you are talking to an AI, and Clidna is not a real person.
• The chat screen header carries a permanent "AI" label.
• During continuing interactions, a reminder card is shown at least every three hours: it reminds you that you are talking to an AI and suggests taking a break.

#8. Contact us

For questions, objections or correction requests about this protocol: support@clidna.com

If you believe a crisis resource on this page is wrong, incomplete or out of date, please write to the same address — we handle those reports as a priority.

This page is published by {{TUZEL_KISI}}.

#9. When this page changes

This page is updated whenever the crisis resource list changes, whenever the behaviour of the crisis flow changes, whenever we launch in a new region, and after the re-verification carried out before each release.