Safety at Textbound

Last updated: [launch date]

You're talking to AI, not a person

Textbound is an interactive story. The character who texts you, Vantar, is an AI character. His messages are part of a work of fiction: on the web, most of them are written ahead of time around the choices you make, and when AI replies are switched on, replies to what you type are written by an AI model. No human writes or reads his replies in real time.

We remind you of this in several places:

• a screen before the story starts
• a label in the chat header ("Vantar · AI character") that's always visible
• a reminder each time you open the story, repeated every 3 hours if you keep reading
• "(AI)" in every notification, once the iPhone app sends them

AI can be wrong. Don't rely on a character for advice about health, money, legal matters or safety.

Adults only

AI companion chatbots may not be suitable for some minors. Textbound is for adults 18 and older.

• Everyone enters a date of birth before the story starts. Anyone under 18 is blocked, and that decision is stored on our server and in your browser so it can't simply be retried. We don't keep the date of birth itself.
• If something in a chat indicates a reader is under 18, we pause the story, lock it, and review it by hand.
• Our ads target adults only, and the iPhone app will be rated 18+.
• All characters are fictional adults.

Our suicide and self-harm protocol

We keep this protocol in place before any character can chat with anyone. It has four parts.

1. Detection

• Every message you type is checked automatically before anything else happens to it, and before it could ever be sent to an AI model. The check looks for expressions of suicidal thoughts, suicide or self-harm, using [classifier name/method, and the evidence base it's measured against; to be confirmed with our clinical reviewer].
• During early access on the web, this check is a keyword screen, a stand-in while the classifier above is chosen and reviewed. A keyword screen misses many ways people express distress. If you're struggling, please contact 988 directly.
• We test the check against a set of example messages, including indirect and slang phrasings, and re-test it whenever we change it.

2. Response: we step out of the story

• If the check fires, the character does not reply in character. Instead, Textbound shows a message from us that says plainly that the character is AI and can't help in a crisis. What you wrote is not added to your story and is not sent to any AI model.
• The message refers you to the 988 Suicide & Crisis Lifeline (call or text 988) and Crisis Text Line (text HOME to 741741), with one-tap buttons, and to 911 for immediate danger.
• The story pauses until you choose to continue. When you do, it resumes on a calm line, and we don't show sign-up prompts in that session.
• Crisis resources are always one tap away in the reader ("Need help? 988") and at /crisis-resources.

3. Prevention: characters don't produce self-harm content

• Our characters never encourage, describe, romanticize or give methods for suicide or self-harm. The scripted story never uses self-harm as a plot device.
• When AI replies are on, every AI reply is checked before you see it. If it fails, you get the scripted story line instead.

4. Records, review and reporting

• When the protocol fires, we record only that it happened and when. We don't store your message text, your account, or your browser in these records.
• Our crisis and character copy is reviewed by [name/credential, or "a licensed mental-health clinician"; to be confirmed].
• Starting in 2027, we'll report to the California Office of Suicide Prevention each year, as California law requires. The report covers how many times we showed crisis referrals and the protocols described here. It contains no personal information.

What characters will and won't do

• In: romance, slow-burn tension, jealousy, morally grey characters, danger aimed at other characters in the story, and kissing.
• Out: sexual content (romance fades to black), coercion or abuse of you, explicit violence, any minor or minor-looking character, age-play.
• Characters don't claim to be human. Ask Vantar if he's real and he'll tell you he's an AI character. They don't give medical, legal or financial advice.

Healthy use

• Textbound is a story with an ending, not a replacement for the people in your life, and not therapy.
• After about 90 minutes in a day, the character encourages you to take a break. Characters also nudge you toward friends and offline life from time to time.
• New scenes unlock on your own story clock, with quiet hours on by default (11pm–8am for evenings). On the web we don't send notifications.

Reporting a message

• Tap Report under any message from a character. The message is hidden right away, and our team reviews every report within 48 hours.
• On the web, characters never message you outside the story page. Muting and blocking a character will come with the iPhone app's notifications.
• You can also email support@textbound.app about anything that worries you.

Your data

• You can see, edit and delete everything the character remembers in What he knows.
• You can delete your chats, his memory, or your whole account at any time. Deletion is immediate in the app and completed in our systems (including backups) within 30 days.
• If AI replies are on, we name the AI provider (Anthropic) and ask for your permission before anything you type is sent to it. It's a paid service whose terms say your data isn't used to train models.
• Full details are in our Privacy Policy.

Contact

[developer name] · support@textbound.app. We reply within 2 business days, and to reports within 48 hours.