Approvals & security — AETHON Docs

AETHON makes administrative changes in your community, so it is designed on the assumption that the AI model can be wrong or manipulated and that member content can be hostile. Authorization never depends on the AI model.

Who can do what#

Every operation has a risk level — LOW, MEDIUM, HIGH or CRITICAL — and every member has an AETHON level: viewer, operator, admin or owner (see operator levels).

Under the default policy LOW and MEDIUM operations run directly when your level allows them. HIGH operations need an approval from an AETHON admin, and CRITICAL ones from the server owner.
Some operations always need an approval and can never run inside an automation: deleting channels or roles, bans and project rollback.
The server owner can change the policy per risk level (/permissions risk) or per tool (/permissions tool); /permissions policy shows the current policy.
Channels and roles protected with /permissions protect cannot be changed by AETHON.
The server can require a different person than the requester to approve.

Checked before every operation#

Right before an operation runs, rules in code — not the AI model — check:

your AETHON level and the risk of the operation;
the Discord permissions of the user and of the bot, server-wide and in the channel;
the role hierarchy, protected resources and roles managed by integrations;
the server's policies, disabled skills and the features of its plan.

Buttons and menus check the user, the server and the current state again on every click.

Approvals#

An approval covers the exact operation and the state of the roles, channels and members involved. Both are checked again right before execution: if either changed, a new approval is needed. Approvals expire if nobody decides them, and an approval given in one project can never be used by another.

What the AI model can and cannot do#

Only the request of the person using /agent is an instruction. Member messages, channel names, files and tool results are treated as data, so a message in your server cannot give AETHON orders.
The model can only propose a plan. The plan then goes through validation, the permission checks and approvals before anything happens.
The model has no direct access to Discord, databases or files — only to AETHON's tools, and only to read-only ones while planning. No tool runs code or commands.
No tool can change AETHON's security policy, operators, approvals or protections. Only people can, with /permissions.

Bans and moderation#

Bans are CRITICAL operations: they always need a human approval, are never allowed in automations and are excluded from progressive moderation. Live moderation only warns, times out or kicks according to the rules you configure.

Never repeated blindly#

When AETHON cannot know whether an operation reached Discord (for example after a timeout), the task is blocked instead of retried. An admin checks Discord and then uses /tasks force_retry if it is safe.

Your data#

Each server's data is isolated from every other server.
Webhook tokens are encrypted and never shown by commands, tools or the audit trail, and tokens and keys are removed from logs.
Uploaded files must be in an allowed format; executable files are rejected and files are never executed.
Restoring a backup never restores the permission policy.
How long project and audit history is kept depends on the plan (see Plans & usage).

Plans and payments#

PRO is activated only by a confirmed subscription for the server. A claim of payment, a message or a command never activates it. Payments are made with PayPal, so AETHON never sees card or bank details.

Good practices#

Keep the permissions requested by the invitation (no Administrator) and keep AETHON's role above the roles it manages but below your staff roles.
Protect critical channels and roles with /permissions protect.
Review /audit history and the Discord audit log regularly.

Reporting a vulnerability#

Please report vulnerabilities privately to the AETHON team through the Support page, and do not share exploit details publicly.