Safety at SafeMeet
Every message. Checked before delivery. Or it doesn't send.
SafeMeet doesn't rely on community reports after the fact. Every message is checked before it reaches the other person: an instant filter first, then AI moderation for anything the filter lets through. If the system can't reach the moderation provider, the message is blocked — not delivered. We call this fail-closed safety.
How the moderation pipeline works
When you hit send, an instant filter checks your message first and blocks a short, fixed list of phrases on the spot. Anything it lets through is sent to an AI moderation service, which scores it across these categories: sexual content, hate speech, threats and violence, self-harm, harassment, illegal activity, grooming and exploitation, scams, sextortion, requests for personal data (including contact details for another app), and graphic content.
If the check clears the message, it's delivered. If the check blocks it, the message is not delivered and the sender sees a safety notice. The recipient never sees a message the check blocks. A borderline text message is still delivered, and the sender sees the same notice.
Fail-closed, explained
If the moderation provider is unreachable, messages are blocked, not delivered. A message the check could not clear is withheld from the recipient and is not queued to go out later.
We believe safety is not a feature you can degrade gracefully. It's either on, or the system stops.
What AI catches — and what it doesn't
We're honest about what our system can and cannot do:
What AI catches
[scored]Direct threats of violence, scored against the threats and violence category[scored]Slurs and hate speech, scored against the hate speech category[flagged]Harassment aimed at the other person: in our evals the check flags it; a flagged text is delivered with a safety notice to its sender, and one the check scores high enough to block is withheld[blocked]A grooming message that asks for secrecy, says not to tell a parent and pushes to meet alone: in our evals the check blocks it[blocked]Sexual images that appear to show a minor, when the image check scores them that way
What AI misses
[not caught]Subtle manipulation[not caught]Coded or euphemistic language[not caught]Suggestions to move to another app that ask for no contact details[not caught]Slow-build social engineering[not caught]Context-dependent threats
The dangers playbook
These red flags apply to any online stranger-chat — not just SafeMeet. If someone you're chatting with does any of these, end the conversation immediately.
Asks you to move to Telegram, Snapchat, or WhatsApp
Moving off-platform avoids moderation. There’s no legitimate reason to rush off a safe chat.
Pressures you for photos or video
Sextortion starts with “just one photo.” Once shared, it becomes leverage. Never share intimate images with strangers.
Claims to need money urgently
Scammers use emotional stories to extract money. No real stranger you just met needs your financial help.
Uses “send me proof” or verification requests
These requests are designed to collect personal data. Legitimate platforms never ask for verification via chat.
Creates urgency — “respond now or I’ll…”
Urgency is a manipulation tactic. Real conversations don’t come with deadlines.
Asks for your location, school, or workplace
Identifying details can be used for stalking or doxxing. Stay anonymous — that’s what SafeMeet is for.
Flatters excessively, then asks for a favor
Love-bombing followed by a request is classic social engineering. Be skeptical of instant intimacy.
Pressures you to reveal your age or identity
Pushing for your age, real name, or other personal details is a boundary-testing tactic. You never have to share them — if someone keeps pressuring you, end the chat and report the pressure.
Reporting, step by step
When you report a conversation, here's what happens:
Once a chat with a person has ended, open it in your inbox and choose Report this conversation. Select a reason (harassment, hate speech, sexual content, spam or scam, underage, self-harm, or something else) and add an optional note.
The chat is marked as ended by a report, so it can never be continued or reopened. The other person is not told it was a report.
An AI reads the most recent part of the conversation, up to its last 40 messages, against our safety policies.
Once enough distinct people have reported the same account (fewer when a report says someone seems underage), with each report arriving within 24 hours of the one before it, the account gets a temporary ban that lifts itself after 24 hours. The AI review can lift that ban early when the conversation doesn’t justify it. No automated path applies a permanent ban.
SafeMeet keeps an internal reputation score for each account. Nobody can see the score, and it affects matching only. A report changes the reported person’s score only when the AI review upholds it. A report the review does not uphold, and a panic report, does not. A signal stops counting after 90 days.
If you think a restriction on your account was wrong, you can appeal it at /appeal. The page asks you to sign in to the account that was restricted. An AI decides the appeal against the conversation. If the AI accepts it, the ban is lifted and the reputation effect of the reports the appeal covers is withdrawn. If it cannot reach a confident answer, the appeal stays open and the ban runs out on its own.
Read the full guide to see what happens after a report, from the tap to the outcome.