---
title: "Confirmation before dangerous actions"
locale: en
url: https://tlg.builera.app/en/docs/concepts/confirmations
verified: 2026-10-24
app_version: "2.0"
---

# Confirmation before dangerous actions

Understand how the app and agents ask for confirmation before dangerous or irreversible actions.


Confirmation gives you a chance to stop before a risky action. It does not replace access permissions, account scope, or the safety gate. A confirmed request can still be rejected because permissions are missing or conditions are not met.

## Types of risk

Each operation has a risk level: read (`read`), write (`write`), or dangerous (`dangerous`). Some write operations become dangerous depending on the data you submit.

| Example | Why it needs attention |
|---|---|
| `messages.delete` | Deletes messages |
| `conversations.clear_history` | Deletes conversation history |
| `accounts.logout` | Logs out the current session |
| `accounts.kill_session` | Terminates a Telegram session |
| `groups.delete_topic` | Deletes a topic and its contents |
| `groups.delete_chat` | Deletes a group or channel for everyone |
| `jobs.redrive` | May repeat an effect |
| `accounts.update_settings` with `limitMode: free` | Removes Safe mode's soft protections |

These are examples, not a complete list. Each operation's [REST reference](https://tlg.builera.app/en/docs/reference/rest) states its risk level and whether confirmation is required.

## How must an agent ask the user?

REST and MCP use the same mechanism. A `409 confirmation_required` response contains `summary` (a description of what is about to happen) and `confirm_token`.

The agent must:

1. Stop.
2. Present the full meaning of `summary` to you.
3. Wait for your approval of that exact action and scope.
4. Call again with the same data, adding `confirm_token`.

The token is valid for **10 minutes**, can be used once, and is bound to that exact request. If the recipient, account, or content changes, the old token cannot be used. Giving an agent a general goal does not mean approving every dangerous action it might take later.

`validate_only: true` checks the request without carrying out its effects.

MCP tool names use the operation code with dots replaced by underscores; for example, `groups.delete_chat` corresponds to `groups_delete_chat`. Tool availability also depends on the connection profile and permission scope.

## Deleting a group or channel

> [!WARNING]
> Deleting a group or channel for everyone is permanent. This is not the same as leaving the group or hiding it from Inbox.

In the deletion dialog, you must type the displayed name exactly. The delete button is enabled only when the name matches. Leading and trailing spaces are ignored, but accents and letter case must match.

## Ownership transfer and two-step verification passwords

> [!WARNING]
> Transferring ownership changes who owns the group or channel. Confirm only after checking the intended new owner.

Only a person can transfer ownership, in the app; AI connections do not have this operation.

You enter your two-step verification password for that specific transfer. The password is encrypted, passed only to the machine assigned the job, then deleted immediately — entered once, used once. Do not put the password in messages to an agent, logs, or API examples.

## Related

- [Dangerous actions](https://tlg.builera.app/en/docs/safety/dangerous-actions)
- [Keeping secrets](https://tlg.builera.app/en/docs/safety/keep-secrets)
- [AI connections](https://tlg.builera.app/en/docs/integrations/ai-connections)
- [The groups.delete_chat operation](https://tlg.builera.app/en/docs/reference/rest/groups.delete_chat)
- [MCP reference](https://tlg.builera.app/en/docs/reference/mcp)