Secure terminal

Fixing what an alert found used to mean leaving Legit AI, finding the server’s address and logging in by hand. The secure terminal opens that shell in the console, next to the alert — and makes every session provable afterwards.

What protects it

• A fresh code, every time. Opening a session asks for the current code from your authenticator app. Each code works once; wrong codes count towards the sign-in lockout.
• Its own key, for two minutes. The terminal never logs in with the monitoring key. For each session Legit AI makes a new key pair in memory and signs a certificate for it that is valid for two minutes, allows a terminal and nothing else (no port, agent or X11 forwarding). The server’s own auth.log names the person and the session from the certificate.
• Only where switched on, only as the monitoring account. An administrator switches the terminal on per server. That adds one line to the monitoring account’s authorized_keys — no root, no sshd change. Removing the line switches it off, whatever happens to Legit AI.
• Recorded and sealed. Everything the screen shows is recorded (AES-256, the same key as other secrets), attached to the alert, and hashed when the session ends. A replay checks the hash and says if the file was altered. Passwords typed at a prompt are never written down.
• A guard on destructive lines. A line that would wipe a disk, power the server off, flush its firewall, delete accounts, cut SSH access or destroy logs is held until you type a code; otherwise it is cancelled. This catches slips and the wrong window — not someone set on getting round it. The recording is the real record.
• Limits. 5 minutes without a keystroke or 30 minutes in all ends a session; one open session per person, three per server, twenty a day.
• Others see it. Everyone in the organisation is told when a session opens; an administrator can watch it live or end it.

Switch a server on

1. Servers, open the server, then its Access tab. The terminal’s state for that server is shown there, beside the list of logins allowed on it.

2. Switch on (administrator, with a code). Legit AI adds the line over the monitoring connection, then logs in with a test certificate. It says Terminal on only when that worked.

To do it by hand, open Switch a server on by hand on the same tab and run the command shown as the monitoring account, then press Switch on so Legit AI tests it.

Open a session

From an alert’s Quick response panel choose Open terminal, or Open a terminal on the Secure terminal page. Pick the server, say why (at least 8 characters — your colleagues see it), and type your code.

Beside the shell you get the alert, read-only checks built from its evidence (who is logged in, the attacker’s address in the auth log, the suspicious process…), the response actions, and the lines typed so far. A suggested check is typed for you; it runs only when you press Enter.

When you are done, End session and, if the problem is fixed, Close the alert as resolved — the session is already on the alert as evidence.

Solo or four eyes

Secure terminal → Policy sets who may act:

	Solo (cloud default)	Four eyes (default on your own server)
Terminal session	opens after your code	waits for a second administrator
Block an address, kill a process, quarantine a file	runs after your code	request, approved by someone else
Isolate a host, harden SSH	your code + the host’s name typed; isolation at most 4 hours	request
Recipes, credential resets, platform accounts	always a second administrator	always a second administrator

Never alone, whatever the policy: your own address, Legit AI’s address, a monitored server’s address, loopback and reserved ranges; PID 1, the listening sshd and systemd. At most 10 direct actions per person in 10 minutes.

Pause now on the same page stops every direct action and every new session, and ends the open ones.

sudo

The session runs as the monitoring account. If it needs sudo, sudo asks for that account’s own password; Legit AI never asks for, sees or stores it.