If you've read anything elsewhere describing AI that scans your video frame by frame, or a hardware fingerprint tracking your device — that's not accurate, and not how Chatzyo works. This page is the real, complete picture, written plainly rather than dressed up to sound more technical than it is.
This isn't a policy choice we made and could just as easily reverse — it's a direct consequence of how Chatzyo is built. Video and audio travel peer-to-peer between your device and the other person's using WebRTC, and that stream never passes through our servers at all. There's genuinely nothing centralized for an automated system to analyze, because the video was never ours to access in the first place.
Some larger platforms with bigger engineering teams have built real-time content scanning into their products. Chatzyo hasn't, and the same peer-to-peer architecture that keeps your video private by default is also what makes that kind of centralized scanning structurally difficult to bolt on later without changing how video is routed. We think keeping video off our servers entirely is the right tradeoff, but it is a tradeoff — it means we depend on people reporting problems rather than catching everything automatically.
Here's the real sequence, with nothing skipped or dressed up:
This is the only trigger. Nothing about your session is reviewed before this point — there's no background monitoring running while you chat.
What's recorded is the category selected, a session identifier, and a timestamp. Not video, not audio, not a transcript — those don't exist to log.
A real moderator on our team looks at the report, with a target of doing so within 2 hours. They decide what action fits, based on what's actually in front of them.
Because Chatzyo doesn't use accounts, restrictions apply to the IP address connecting to the platform — temporary for less severe issues, permanent for serious ones, exactly as laid out in Section 7 of our Terms of Service.
Unlike video, text messages do pass through our server briefly to get delivered to the other person. A simple filter checks for patterns like phone numbers and email addresses and blocks the obvious attempts to share them — this is pattern-matching on text, not a person reading your conversation, and the messages themselves aren't kept once the session ends. It's a narrower, more limited system than people sometimes assume, and we'd rather describe it accurately than imply it's smarter than it is.
You might come across claims elsewhere — a specific accuracy percentage, a precise share of bad actors caught before they ever reach another user — stated with confident precision. We don't have numbers like that to publish honestly, because the system those figures would describe doesn't exist here. Chatzyo is a young platform without a year of audited moderation data behind it, and a made-up statistic would be worse than no statistic at all. If we ever have real, meaningful numbers to share, they'll go in our moderation transparency report — not invented for a help article.
Because there's no automated system catching problems in the background, the report button carries more weight on Chatzyo than it might on a platform that also runs content scanning. If something happens and no one reports it, there's no second layer quietly catching it afterward. That's worth knowing both ways — it's a real limitation, and also the reason reporting promptly genuinely makes a difference here, rather than being a formality layered on top of something else doing the real work.
For what to actually do if something goes wrong, see our guides on reporting a user and handling harassment. For the full rules the moderation team is actually enforcing, see Community Guidelines.