Loading Gatelion
Skip to content
GatelionBetaStart free

Security

How Gatelion keeps agents within your rules

Gatelion exists to stop AI agents from doing what you didn't allow. This page explains how, part by part, so your security team can judge it. Gatelion hasn't had an external security audit or certification yet.

The parts, and who trusts whom

Organizations are kept apart

Every query the Control Plane makes is limited to the signed-in person's organization, and the limit fails closed: a query without it returns nothing. Another organization's resources answer as if they didn't exist. Gatelion's own back office sees organizations' status, usage counts and activity dates, never their agents, approvals or data.

What agents can do

Approvals

The Gateway's copy of your configuration

Gateways work from a snapshot of the configuration that the Control Plane signs. A Gateway checks the signature against the key it pinned at registration, refuses older or foreign snapshots, and keeps only verified ones. If it loses contact, it keeps deciding for a limited time, set per environment, then refuses writes, and later reads too.

Calling your systems

Your systems' secrets

Signing in

The audit trail

Requests, decisions, approvals, results and administrative changes are recorded as events. Each environment's events form a hash chain, so changing, removing or reordering one breaks it. The console verifies the chain on demand. The application's own database account can add events, but can't change or delete them.

In transit

The console, the Control Plane and the hosted Gateway are served over HTTPS. A Gateway in your network serves HTTPS with your own certificate. If one runs on plain HTTP, the console marks it, because agents' tokens would cross your network unencrypted. See Install a Gateway.

Known limits

Reporting a vulnerability

Write to contact@gatelion.si with what you found and how to reproduce it. Please give us a chance to fix it before you share it. We'll confirm we received it and keep you updated.