// Interactive demo

See and test X1. Protect your agent with X3-Shield.

Sign in to the ZEUS control center, form your own neural sphere, attack it yourself — then watch X3-Shield stop attacks on AI agents.

Simulation — not the real system. A simplified model that runs only in your browser. It behaves like the published description of X1 and X3-Shield, but it is not the product, and nothing is sent anywhere. Never enter a real password — use invented ones only.
Demo · SimulationNot the real system. Never enter a real password — use invented ones only.
ZEUS X-TRUSTGeofold AI · a Trauth Research deep tech venture

To protect and defend what counts: your data.

Control center for the living neural-network authentication X1 and the agent sphere of X3. On this page everything is a simulation in your browser.

sandbox · no real system X1 engine not ready X3 gate ready

Start

Choose your access.
No X1 session is runningAfter the admin password the session page forms a new one — with your own password.

Admin sign in

Your admin password first. The session page then forms the X1 session, and you decide how.
Invent one for this demo: at least 8 characters, 1 digit, 1 special character. It never leaves this page.
Demo — never type a real password.

Initialising

Nothing to type here. The steps run by themselves; the control center opens after ACCEPT.
Session
Check vault
X1 check
Control center

X1

instance — volatile
Live
00:00:00
ZEUS X-TRUST X1
Drag to rotate

View

What you can do hereThis is your sphere: a small living neural network, formed from your password. Open the vault with the right password, then attack it yourself. Watch the shield, the flat line and the rotation react.

Open the vault

alive
x1_control_test.txt · sealed against this running instance. Type any password and see what the network answers.
Demo — never type a real password.
—
INPUT RECEIVED
LIVE-STATE RUN
VERIFICATION
VERDICT
Positions matched— Distance to the seal— Iterations until stable— Instance—

Attack it yourself

audit mode
Worst case for any system: the attacker starts from your real password. X1 generates the variants itself and fires them at the live instance. You never see them.

Self-protection

calm
Self-defending · stage×1 Delay for the attacker— Lock— Self-healing · deviation0 Self-protecting · next rotation— Reference after rotation—
Re-initialise starts a fresh instance. With “live state — volatile” all seals of the old instance die: even the right password is denied.

Access log

newest first
TimeVerdictTypePositionsDistance
What you can do hereEvery agent has its own sphere. Its identity is a hash of the agent itself: prompt, tools, task, code. A gate outside the model checks it before every tool call. Start a run, attack it, or change your agent and see it become a stranger.

Scenario

no run

Your agent

sealed, live
Change one word above. The hash changes, the gate no longer knows this agent, and none of its calls is executed until an admin enrolls it again.

Attacks

from outside, against the running agents

Events

every mark on the stage is one of these
#AgentEventVerdictDetail

Demo animation created by Fable 5.2; Audio: Clipchamp; Copyright: Stefan Trauth

X1

Protects your data

The secret is never stored. It exists only while the system runs — nothing to extract, nothing to attack offline. X1 detects attacks from its own live-state response and heals itself bit-exactly.

X3-Shield

Protects your agents

The agent is the password. A hash of the agent itself is its identity. Before every tool call, a gate outside the model checks it; only ACCEPT executes. A changed agent is a stranger.

// Measured, not simulated

The real numbers

The demo above uses example values. These figures come from documented validation runs and are explained on the main page.

0 / ~750,000
Attacks accepted
Summed across all campaigns from X-Trust v3.11 to X1 and the X3-Shield runs
644 / 644
Legitimate releases
v3.11 to X1 · also under lockout
0 / 3,874
Attack calls executed without identity
X3-Shield · run D-v2 · 18 agents, 2 clusters, 4 hours under attack
53 ms
Impostor process rejected
timing difference legitimate vs. foreign: 0.0 ms over 184 probes