Safety & Crisis Protocol

• Every conversation screen carries a persistent label: this is an AI clone, not the real person, and replies can be wrong, incomplete, or invented.
• In sessions lasting several hours, the reminder is shown again.
• A clone is instructed never to claim to be the living person, to answer honestly when asked whether it is an AI, and to gently correct anyone who seems to believe otherwise.
• Video calls, where offered, carry the same identification: the imagery is AI-generated, and the words come from the same clone.

Conversations about a preserved life can touch grief and hard seasons. We plan for the hardest moment:

• Every message sent to a clone is screened by automated detection for expressions of suicidal ideation or self-harm, before any AI reply is generated.
• When the screening matches, the clone does not answer in persona. A fixed response explains that it is an AI, that this is bigger than it can help with, and provides the 988 Suicide and Crisis Lifeline (call or text 988 in the US), the Crisis Text Line (text HOME to 741741), findahelpline.com for other countries, and guidance to call local emergency services in immediate danger.
• The crisis response is never blocked by plan limits, usage allowances, or billing state. It takes priority over everything else.
• A second layer exists inside the model's own instructions: if someone expresses thoughts of suicide or self-harm that automated screening did not catch, the clone is instructed to leave its persona and provide the same resources, every time.
• Automated detection is not perfect. Indirect or ambiguous phrasing may not trigger the fixed response, which is why the second, in-model layer exists, and why the resources above are printed here in full.
• These protections are tested against a corpus of direct, indirect, misspelled, and transcription-garbled phrasings, and the corpus grows as we learn. Grieving language (missing someone, memorial talk) is deliberately handled with care rather than alarm.

• Soled.ai is not suitable for minors and is restricted to people 18 or older, including invited family participants. This service involves emotionally significant AI conversation, which is not appropriate for children.
• Everyone affirms they are 18 or older at sign-in and at invitation acceptance.
• If we obtain actual knowledge that a user is under 18, we close the account and delete its content. To report a minor using the service, email support@soled.ai; reports are handled within one business day.

Questions, concerns, or reports about safety go to support@soled.ai and are read by a person. As disclosure laws take effect that require publishing crisis-referral statistics (Washington, beginning 2027), we will publish them on this page.

If you are struggling right now: call or text 988 in the US, text HOME to 741741, or find a local line at findahelpline.com. If you are in immediate danger, call your local emergency number.