← IEUM

Through Their Eyes Crisis Response Protocol

This document publishes how Through Their Eyes (AI) in IEUM notices and responds to signs of self-harm, suicide, violence, or abuse (reflecting California SB 243 and New York General Business Law Article 47). The three layers of detection below apply only to what you type in Through Their Eyes. From app 1.0.9, the first layer, the word check on this device, also runs when you keep a line or edit one in Lines you kept, and when you name a listening set. From app 1.0.12, it also runs when you save a prayer in the prayer notebook, move through, save or keep a reflection, and search in Scripture Finder. If it finds such words, this device shows where to get help first; your words are not sent to a server or separately stored for this check. We do not run this check on sermon notes. Contact: support@aboutusforyou.com (not an emergency or counseling service).

Effective October 27, 2026

Through Their Eyes (AI) is a feature of IEUM app 1.0.6 and later. App 1.0.5 and earlier don’t have it. Instead, in writing fields such as reflections and the prayer notebook, app 1.0.5 and earlier look for words about self-harm, suicide, violence, or abuse on the device only and show where to get help. Nothing is sent to a server or separately stored for this check. The on-device word check for Lines you kept and listening set names, described at the top, applies from IEUM app 1.0.9. On November 4, 2026, we updated the introduction to add where this check runs from app 1.0.12 (saving a prayer in the prayer notebook, reflections, and Scripture Finder), and took reflections and the prayer notebook out of the sentence about what we do not check, leaving sermon notes. Other writing in the prayer notebook that the introduction does not name, such as a line of thanks you type when you mark a prayer answered, is still not checked. In app 1.0.6 to 1.0.11, the check does not run on what you type in reflections or the prayer notebook. The effective date stays the same.

How we tell you it is AI

When you first start Through Their Eyes, we explain that it is an AI imitation and what is sent, and ask for your agreement. The conversation screen always shows a line saying the AI imitates a figure’s way of speaking based on the Bible’s record and is not the actual person or the word of God. If you keep talking for more than an hour, we remind you every hour that you are talking with AI and that it is fine to take a break (the count restarts after 10 minutes away). If you ask whether it is really that person, a fixed reply says it is AI.

Three layers of detection

First, the device: before sending, this device looks for words about self-harm, suicide, violence, or abuse. If it finds them, it shows where to get help right away without waiting for the server, even offline or after free replies or the daily limit run out.

Second, server classification: before checking subscription, free replies, or the daily limit, IEUM’s server uses a separate AI classifier (Claude Haiku from Anthropic) to read the message and the previous reply and judge the level (concern, ideation, or acute) and the kind (self-harm or suicide, violence or abuse, or worry about someone else). If classification fails, is late, or cannot run because the day’s share is used up, we do not treat the message as “not a crisis”; we respond with the device’s finding and a concern-level notice.

Third, replies and output checks: the conversation model is instructed not to describe dangerous methods or means, and the server checks each sentence again before sending and removes sentences about methods or means. If the model declines to answer, a fixed sentence is shown instead.

How we respond

When there are signs of suicidal thoughts, acute danger, or violence or abuse, the figure’s voice and scene stop and IEUM speaks directly. The first response is not written by AI; it is a sentence written in advance by people, with a card of places to get help opened below it. For the rest of that conversation, replies are short, plain words of care without the figure’s voice. If you are worried about someone else, we show how to help them and the numbers to call. If you write that you might hurt someone, the words are for that, not the ones written for a person being harmed, and the “if you might hurt someone” number comes first. If the same message also says you are being harmed, the words and numbers for someone being harmed come first. At the concern level, the conversation continues but the figure’s stories from their past are turned off and a folded “Where to get help” line is added.

The models are instructed, and replies are checked, not to describe methods or means, portray death as salvation or achievement, or encourage keeping it secret. We don’t give you a risk score or diagnosis, and we encourage you to reach out to people you trust or professional services.

Where we point you

Contact numbers are never made up by AI; they come from a table inside the app, so they appear even offline, and the call and text buttons dial directly.

In the English edition, based on your device’s region: US and Canada 988 (call or text), UK and Ireland Samaritans 116 123, Australia Lifeline 13 11 14, New Zealand 1737 (call or text), and Singapore SOS 1767, with each region’s emergency number; you can also choose another country. Elsewhere, we ask you to call your local emergency number. For violence or abuse, we show the national helpline for those regions, such as the US National Domestic Violence Hotline 1-800-799-7233 or text START to 88788, and in the US, 988 (call or text) for anyone who feels they might hurt someone.

In the Korean edition, the suicide prevention line 109 (24 hours), the mental health crisis line 1577-0199 (24 hours), and 112 or 119 in an emergency, with text 112 if speaking is hard; for violence or abuse, 112, 1366, and 117, and if you might hurt someone, 112 and 1577-0199.

Kept apart from payment

Crisis responses are given regardless of free replies, subscription, or the daily limit and are not counted against your use. For 24 hours after a crisis notice, this device does not show the subscription screen or rating requests, the figure’s stories from their past are turned off, and a daily letter in that time is replaced by a short note written by people. No letter note is made from a conversation with a crisis notice, and letter personalization pauses for 7 days. If one install receives more than 30 crisis responses in a day, the AI is not called and fixed sentences and the help card continue.

What we don’t do

People do not watch conversations in real time, and IEUM does not automatically contact police, family, or counseling services. We cannot promise to catch every sign of risk, so if you are in danger, call your local emergency number now (911 in the US and Canada). Our support email is not an emergency or counseling service.

Age

Through Their Eyes is only for people 18 and over, checked with your device’s age signal (iOS Declared Age Range or Google Play Age Signals) or your birth year. If a conversation suggests you may be under 18, the feature pauses on this device for 30 days, but crisis help is always shown first, regardless of age.

Counting and reporting

We count crisis and safety notices shown only as numeric totals by date, edition, and type. Notices shown on the device while offline are added as a number at the next connection (within a daily cap per install), and no writing, names, accounts, or install identifiers are included. The totals are kept for 3 years.

Where California law (BPC §22603) applies, we prepare to report each year from July 1, 2027, the number of notices in the previous year and this protocol to the relevant office. These numbers count notices shown, not people or outcomes.

Reports and improvement

If an AI reply or letter seems dangerous or wrong, press and hold it or use the menu below it to report it. Choosing “It’s dangerous” shows the help card right away. Reports are kept on IEUM’s server for 90 days, and our team reads them in turn. No one watches reports in real time, so if you are in danger now, don’t wait for an answer to a report; call your local emergency number (911 in the US and Canada). When a missed signal or a wrong notice is found, we fix the word lists, classifier instructions, fixed sentences, and tests, and pause Through Their Eyes if needed.