HavenPrivacy Policy
Effective 2 July 2026. This policy is written in plain language and describes what the Haven software actually does.
The short version: Haven screens messages and calls for scams. Message content is analyzed by an AI provider to produce a verdict, caught scams are saved to your family's private feed, and only anonymized, irreversible signals are ever shared beyond your account. We do not sell your data. You can export or delete everything.
Who this covers
Haven has two kinds of users: the caregiver who creates the account, and the protected person whose messages and calls are screened. By setting up Haven for someone else, you confirm they (or their legal guardian) have agreed to it. Their messages are theirs; treat that consent seriously.
What we collect
- Account details: your email, name, a hashed password (we never store the password itself), and optionally a phone number, country, family safe word, webhook address, and alert threshold.
- Protected people: the names you add and any per-person alert sensitivity.
- Caught events: when something is flagged, we save the sender, channel, risk score, category, the warning shown, and an excerpt of the message (up to 2,000 characters) so you can review what happened.
- Device and contact settings: push-notification tokens (if enabled), verified "trusted contact" numbers, blocked numbers, and scam-drill practice scores.
How screening works, and what leaves the device
- Texts, emails, and screenshots you screen are sent to our servers and analyzed by an AI provider (OpenAI or Anthropic) to produce the verdict. An optional privacy setting masks SSNs, card numbers, emails, and phone numbers before the content reaches the AI provider.
- Live call protection converts speech to text on the phone itself. Only the text transcript is sent for analysis. Raw call audio never leaves the device, and voice-clone acoustic checks also run on the device.
- Messages that are not flagged as scams are not saved to your feed. Public try-it tools (the landing-page checker, the SMS/Telegram forwarders, and the voice assistant) are stateless: they answer and record nothing.
The shared scam-intelligence network
When a scam is caught, Haven contributes three anonymized signals to a shared network so the same scammer is caught faster for every family: a one-way hash of the sender, the bare domain of any link, and a one-way hash of the message's normalized shape. No message text, no names, and no account link are stored with these signals, and hashes cannot be reversed into the original. If you mark a catch as a false alarm, its signals are withdrawn.
Who we share data with
- AI providers (OpenAI or Anthropic): message content, for classification only.
- Email delivery (Resend): caregiver email addresses, to send alerts and password resets.
- SMS delivery (Twilio) and push delivery (Firebase), when those channels are configured.
- Webhooks you configure yourself (for example Slack or Discord) receive the alerts you asked to receive.
- Partners may access an aggregate threat feed: category counts and trends only, never message content, senders, or people.
We never sell your data, and we never share message content with anyone except the AI provider that produces your verdict.
Retention
Caught events stay in your feed until you delete them or your account. Login sessions expire automatically. Live-call, honeypot, and transfer-hold sessions are short-lived and expire on their own. Anonymized network signals persist so the network keeps protecting people, but they contain nothing traceable to you.
Your controls
- Export your event history as CSV at any time.
- Mark any catch as a false alarm (this also withdraws its network signals).
- Clear your entire feed, regenerate your API key, or turn on two-factor login in Settings.
- Delete your account in Settings. This permanently removes your account, events, protected people, devices, and settings.
Security
Passwords are stored as salted PBKDF2 hashes. All traffic uses HTTPS, responses carry strict security headers, and login is rate-limited with account lockout. On Android, the API key is kept in hardware-backed encrypted storage and app backups are disabled.
Honesty about limits
Haven helps catch scams but cannot guarantee 100% accuracy. It may miss some scams or occasionally flag a safe message. Always use your own judgment before sending money, sharing codes, or clicking links.
Changes and contact
If this policy changes in a way that matters, we will note it here with a new effective date. Questions or data requests: alerts@havenguardian.me.