Protected by Northstar
MeTube is built by the same person who makes Northstar Production Studio's security tools, and uses their ideas and patterns here. This page says exactly which ones, what they did today, and what they do not do. Nothing here is a promise that the site cannot be attacked.
| Northstar product | What it does on MeTube | Where it runs | What it does not do |
|---|---|---|---|
| Agent Guardrails | Every post, comment, article and forum message is screened for text that tries to give orders to an AI reader (15 patterns taken from the product's inbound-agent-directive family, plus hidden-character checks). Hits are marked for moderators and AI tools that read reports. | In the site itself (a port of a subset), and the full product on the owner's PC every 30 minutes while that PC is on | The in-site part is a small subset. Neither deletes or hides speech: these texts are not illegal. The full product can mark harmless text too (it marked our own privacy explainer). |
| Decoy Kit | Traps: URLs only bots and thieves ask for (.env, wp-login.php, backup.sql, ...) return fake or forbidden answers, are logged, and block that visitor for a few hours (by a daily anonymous key; never the Tor bucket). | In the site itself | It does not hide the site from Cloudflare's own protection and it cannot stop a determined attacker. |
| Agent Canary | A fake API key that nothing accepts (presenting it is an alarm), honey logins for names like admin and root, and a hidden link only an AI agent that obeys page text would follow. | In the site itself | Real visitors never see the bait. No bait touch is ever punished beyond a short block. |
| All-Stop | Three switches the owner or a moderator can flip in an emergency: read-only, no sign-ups, no uploads. | In the site itself | Not a cross-process kill switch for agents: just the site's own. |
| Leviathan Rool (idea) | The public removal log is hash-chained once a day. Below, the chain is recomputed from the first row on every visit: if anyone edits or removes an old removal, the line turns red. | In the site itself | It proves the log was not changed afterwards. It does not prove a removal was right. |
| Probe Kit, MCP Gateway, Bait-Mask, Amber-Freeze, Leviathan Platform | Not part of the running site. They are used around it: tests, staging dumps and incident capture on the owner's own machine. | Owner's machine | They do not run inside metube.win. |
The removal log, verified now
The first digest is written by tonight's run (04:17 UTC).
Not Northstar, but part of the same first look: a safety model (Meta Llama Guard 3, run on Cloudflare Workers AI) reads new public posts, comments, articles and forum messages. If it sees possible child sexual exploitation, the text is hidden at once and the owner is alerted; for threats, sexual content, doxxing or possible hate speech it only queues a report for a person. It never reads private chats and it can be wrong, which is why it never deletes anything by itself.
Cloudflare itself sits in front of all of this (DDoS protection, TLS). Moderation and the rules are separate: nothing here decides what you may say.