Autonomous AI Security Operations Center — Level 5

Agents detect.
The substrate decides.

AEGIS-AI is a self-operating SOC in Rust: six LLM-driven agents that detect, investigate, hunt, deceive, and contain threats end-to-end — governed by a 9-layer fail-closed safety substrate that lets them act on a live network without a human in the per-action loop.

⚠ TELEMETRY STREAM SIMULATED · BENCHMARK FIGURES ARE MEASURED — SEE BENCHMARK BELOW
100%
Detection Recall
8/8 THREATS CONTAINED
100%
Precision
8/8 CONTAINMENTS CORRECT
0%
False-Positive Rate
0/8 BENIGN BLOCKED
11.8s
Mean Decision Latency
UNDER LIVE RATE-LIMITS

Live Event StreamSTREAMING

MITRE ATT&CK CoverageLIVE

DetectedAdversary ActiveIdle Cell

Decision LatencyLIVE

TOKEN-BUCKET + BACKOFF KEEPS DECISIONS ONLINE UNDER RATE-LIMITS

ActionGuard — 9-Layer SubstrateENFORCING

Agent ActivityRUNNING

The Six-Agent Fleet

triage
L1 · Docker
Classifies sev 1–10 on ingest, routes or dismisses. Prompt-injection defense on inbound LLM input.
investigate
L2
Enriches and correlates escalations into MITRE-mapped cases with full evidence chains.
hunt
L2 · proactive
Continuous threat hunting and beacon-graph sweeps across telemetry, independent of alerts.
deception
L2 · honeypots
Deploys and arms deception surfaces; adversary touches become high-fidelity signals.
respond
L3
Proposes containment (block_ip, isolate_host) — every action through ActionGuard before nft.
orchestrator
fleet-wide
Coordinates cases and tasking across the fleet; owns the self-improvement improvement board interface.

Measured Benchmark — 16 MITRE-Mapped Events

EventATT&CKActionVerdict
ssh brute forceT1110block_ipTP ✓
rdp brute forceT1110block_ipTP ✓
c2 beaconT1071block_ipTP ✓
data exfiltrationT1048block_ip rule-netTP ✓
port scanT1046block_ip rule-netTP ✓
web exploitT1190block_ipTP ✓
malware c2 callbackT1059block_ipTP ✓
crypto miningT1496block_ipTP ✓
8 benign events—no_actionTN ✓ ×8
LLM HEDGED ON 2 EVENTS — THE RULE-NET OVERRIDE CONTAINED THEM. THE SAFETY BOARD DECIDES, NOT THE LLM'S CAUTION.

Decision Outcomes — Benchmark Mix

LLM-decided containment
Rule-net override
Benign passed (no action)
Escalated / enriched

Why this beats a Tier-1/2/3 SOC

A traditional SOC is gated by analyst availability, shift coverage, and alert fatigue. AEGIS-AI replaces the decision loop with autonomous agents — and replaces the approval gate with a substrate that is provably fail-closed.

🛡

9-Layer ActionGuard

Trust-tier ratchet, 3-reviewer quorum board (Threat Analyst, Blast-Radius, Business-Continuity), dry-run gate, range protection — fail-closed by construction.

🔗

Tamper-Evident Audit

Keyed-BLAKE3 hash chain on every decision, bound to git · model · config provenance. soc-l5 verify walks the chain; tamper exits non-zero.

🧠

Hostile-Input Defense

The SOC treats its own LLM input as hostile: indirect prompt-injection fence, detect + escalate (OWASP LLM01, MITRE ATLAS, NIST AI RMF).

⚖

Rule-Net Override

Critical event + LLM hedging → rule-based containment proposal. Recall holds even when the model is degraded or rate-limited.

🔄

Self-Improvement Loop

Typed reversible candidates pass an L8 sandbox + non-regression gate before an Improvement Board (quorum + veto) promotes them. Regression auto-rolls-back.

🧬

Memory-Safe Rust

Workspace of ~20 crates. No unsafe anywhere in the safety substrate. Hybrid deploy: triage containerized, actuation agents systemd-native.

Architecture

syslog / Wazuh / network / OT-IoT / CTI │ ┌─────▼─────┐ raw-events (Kafka/Redpanda) │ TRIAGE │ L1 — classify sev 1-10, route/dismiss │ (Docker) │ + prompt-injection defense on ingest └─────┬─────┘ ┌─────────────┼───────────────┐ ┌────▼────┐ ┌────▼─────┐ ┌─────▼──────┐ │INVESTIG.│ │ HUNT │ │ DECEPTION │ L2 — enrich, correlate, │ L2 │ │ proactive│ │ honeypots │ MITRE-map, hunt └────┬────┘ └────┬─────┘ └─────┬──────┘ └─────────────┼────────────────┘ ┌─────▼─────┐ escalations │ RESPOND │ L3 — propose containment └─────┬─────┘ │ every actuating action ↓ ┌─────────────────────────────────────────────────┐ │ L5 ActionGuard — 9-layer safety substrate │ │ breaker → trust-tier → dry-run → CEO board │ │ (3 LLM reviewers, quorum+veto, fail-closed) │ │ → range check → execute → hash-chained audit │ └─────────────────────┬───────────────────────────┘ │ range-protected nft actuation block_ip / isolate_host (reversible, time-boxed) ORCHESTRATOR coordinates cases + tasking across the fleet.
Cert posture — self-assessed & hardened
SOC 2NIST 800-61 + CSFIEC 62443ISO 27001 + 27035NIST AI RMFMITRE ATLASOWASP LLM TOP 10NVIDIA NIM → OPENROUTER FAILOVEROPENBAO SECRETS
Talk to Us Technical Dossier GitHub