Moderation in Ansible

What we remove, how it works technically, and what this system does not have yet. Without numbers we do not measure.

What is not allowed

Public surfaces — channels, supergroups, sticker sets, bots — must not carry calls to violence, child abuse material, or trade in prohibited goods. This applies to the public part: one-to-one conversations and secret chats are not reviewed, and technically cannot be — in a secret chat the server has no key.

How a signal arrives

• User reports — from the client, via messages.report and its counterparts for channels and profiles. A report enters the queue with the object and the reason attached.
• Review by an administrator — a person decides, through the admin panel. Every action is written to an audit log: who, what, over which object, from which address. The log is denormalised on purpose — an entry has to stay readable after the administrator's token is revoked.

What we can do technically

• Freeze an account — a frozen account cannot perform state-changing actions; the check sits at the single RPC entry point, so it covers all three transports at once.
• Hide or remove a public object — a channel, a sticker set, a bot.
• Revoke an application — every API consumer holds an api_id issued through my.ansible.su. A banned flag disconnects it from the network within half a minute, and it is a lever separate from accounts: an abusive client is switched off without touching its users.
• Revoke an invite link — a revoked link stops opening, and the link page answers it exactly as it answers one that never existed: otherwise it would become a way to enumerate hashes.

Lines we do not cross

We do not read private conversations and do not scan their contents. In secret chats that property is mathematical rather than a promise: the server relays bytes without holding the key. For cloud chats access is technically possible, and saying so plainly matters more than claiming otherwise — but it is not used for moderation, because what moderation is about is the public part.

What the server sees either way is metadata: who exchanged messages with whom and when, and how large they were. That is a property of any delivery network.

About statistics

We have no public dashboard of numbers yet. The layout and the charts on this page are ready, but there is no data source behind them — we do not publish counts of removed and blocked material until we are confident they are counted the same way from one month to the next. Putting arbitrary numbers here would be worse than having none.

How to report

From the app — "Report" in the menu of a message, a chat or a profile. For anything a report does not solve, see the FAQ.