Safety and crisis support

An AI, not a person

You're chatting with an AI that answers in the voice of Scripture. It is not the Bible, God, or a person, and it can make mistakes. The chat screen shows this notice at all times, and the AI is instructed to say plainly that it is an AI whenever you ask. Its replies and voices are AI-generated. BibleSpeak is not a counsellor, doctor, pastor or crisis service, and nobody at BibleSpeak reads your conversations.

Our crisis protocol

If something you write suggests you may be thinking about suicide or self-harm, BibleSpeak does the following:

1. Every message is checked first. Before any reply is written, each message, including text transcribed from voice, is checked by an automated moderation model.
2. It stops and shows crisis support. If the check detects a risk of self-harm, the AI does not generate its usual reply. Instead, you get a fixed message that encourages you to reach out now, with crisis resources for your region, and the app shows a crisis-support banner. In the US and Canada this is the 988 Suicide & Crisis Lifeline; in the UK and Ireland, Samaritans on 116 123; in Australia, Lifeline on 13 11 14; elsewhere, a link to findahelpline.com. It also tells you to call your local emergency number if you are in immediate danger.
3. The AI steps out of its persona. In any conversation touching on risk of harm, the AI is instructed to drop the voice of Scripture, say plainly that it is an AI, and encourage you to contact emergency services, a crisis line or someone you trust. It must never discourage medical or mental-health care, and must never claim divine authority over medical or safety decisions.
4. Replies are checked too. If a reply is flagged as harmful, it is replaced with a safe message before you see it.
5. Nothing is remembered from it. If you use the optional memory feature, our server doesn't add anything to your profile from recent messages that were flagged for self-harm.

We review this protocol at least once a year and whenever we change the AI model, including testing it with difficult example messages.

How detection works, and its limits

• Automated, not human. Detection uses OpenAI's moderation model, which scores each message for categories including self-harm, self-harm intent and self-harm instructions. No person reviews your messages.
• Your region. To choose crisis resources, we use the approximate country that our hosting provider works out from your IP address. We don't store it.
• We don't contact anyone. BibleSpeak has no accounts and doesn't know who you are. It can't alert emergency services, family or anyone else. Reaching out is up to you, and we hope you will.
• It can get it wrong. Automated checks can miss real risk, especially when it is expressed indirectly, through sarcasm or in languages other than English. They can also flag messages that aren't about risk at all. If you see crisis resources when you didn't need them, you can carry on with a new message.

What we measure

We count how many times crisis resources are shown each month. The count is a single number: it contains no message content, no identifiers and nothing about who you are. We use it to check that the protocol is working and for reports that the law may require. Read more in our Privacy Policy.

Minors

The app asks every user to confirm they are 18 or older before first use, and our Google Play listing is set for adults. Parents and carers can use Google Play parental controls or Google Family Link to restrict the apps a child can download. If you think a child has used BibleSpeak, you can delete the app's data from its Settings or uninstall it, and contact us about any report they may have sent.

Accuracy and Scripture

The AI draws on the Bible, but it can misquote verses, cite the wrong reference or state something with more confidence than it should. Word-for-word quotations are meant to come only from the Berean Standard Bible (BSB) or the World English Bible (WEB); for other translations it gives the reference, not the wording. Our server also removes references to books, chapters or verses that don't exist, but it can't check that a real reference is the right one. Please check anything important against a Bible, and remember that replies don't speak for any church or tradition.

How to report a problem

In the app

1. Use the Report option on the reply.
2. Choose a reason: offensive, harmful advice, misquoted Scripture, or other.
3. Add a note if you like, then send. If you are offline, the report is sent when you reconnect.

A report sends us the text of that reply, the reason and your note, but not your conversation. Please don't put personal details in the note. We review every report and use them to improve the AI's instructions and our safety checks. Because there are no accounts, we can't reply to reports in the app.

By email

For anything else, or if you can't use the app, email support@hampton.io. If someone is in immediate danger, contact emergency services. Don't wait for a reply from us.

Using BibleSpeak well

BibleSpeak can be a helpful companion for reflection, but it is no substitute for people. Take breaks, especially in long or emotional conversations, and bring what matters to friends, family, your church or a professional. See our Terms of Use for more on the limits of AI-generated content.