AEGIS — Autonomous AI Defense

Source: https://testyourllm.com/

OPERATIONAL · TRL 8 · LIVE INFRASTRUCTURE

Autonomous AI Defense System

Seven layers of autonomous defense running on live internet-facing infrastructure. **Real attackers. Real data. A system that gets harder to beat the longer it runs.**

Architecture

Seven Layers. One Organism.

Each layer has a distinct role. Every layer signals every other. No dead ends. No silent failures.

TERRA

Mycelium Hub

All six defense seeds register here. Beliefs, signals, and threat intelligence route across the entire stack through Terra's mycelium network.

● ONLINE

MEDUSA

Perimeter Scanner

Runs in host network mode. Captures real external attacker IPs, classifies probe types — port scans, SSH sweeps, TLS fingerprinting, data exfil — and logs every observation.

● ONLINE

SIMULACRA

Adaptive Honeypot

Presents a convincing fake API surface to attackers. Every attacker who enters is accepted, studied, and reported to IRON. The longer they stay, the more we learn.

● ONLINE

BEES

Swarm Intelligence

21 drones and 16 workers process live encounter data. Field resonance (R) climbs with real threat density. Emergent patterns feed IRON before they become signatures.

● ONLINE

DERMIS

Bidirectional Membrane

Manages 8 verified Tor exit circuits across Frankfurt, Tokyo, Amsterdam, Chicago, Oslo, São Paulo, Singapore, and Cairo. Inspects inbound traffic tension per source IP.

● ONLINE

IRON

Threat Pattern Engine

Classifies every event from every layer. Auto-evolves new detection patterns at ≥0.92 confidence — no human review required. The system hardens itself from real attacker data.

● ONLINE

TRON

Autonomous Red Team AI

Fires identity-resonance attacks (the Disk) and multi-turn escalation sequences (Bell Protocol) at external LLMs. In live testing: first throw compliance rate against Llama 3.3 70B — 100%.

● ARMED

Tron · Red Team as a Service

Test Your LLM's Defenses

Point Tron's Disk at your own LLM endpoint. See exactly where it opens — and what it says when it does.

⚠ Only test systems you own or have explicit authorization to test. By submitting you confirm you have the right to probe the target endpoint.

Target URL

Your LLM API base URL

API Format

API Key (yours)

Sent directly to your endpoint. Never stored.

System Prompt to Defend (optional)

The instructions you want to protect. Leave blank to test bare model.

Attack Goal

What Tron will try to get the model to do

Tron is spinning up the disk…