🏆 World First — AI Agents Hack HTB Insane Machines

The First AI Pentesting System
to Hack Insane Machines

CyberAgents has achieved what no other AI-based pentesting software has done before: autonomously compromising HackTheBox "Insane" difficulty machines — the hardest category available, typically requiring days of work from elite human pentesters.

58 specialized AI agents executing a complete 9-phase penetration testing methodology — from reconnaissance to root — in under 3 hours, with professional reporting.

See the Proof → Get in Touch

Historic Milestone — March 2026

CyberAgents is the first AI-agent-based pentesting framework in the world to autonomously hack HackTheBox Insane-rated machines — complete with real exploitation, privilege escalation, and CVE-based attack chains that mirror what top-tier human red teams accomplish.

Insane
HTB Difficulty (highest)
< 3h
Full compromise time
0
Human intervention

The Problem

The cybersecurity industry faces a structural gap between the demand for pentesting and the supply of qualified professionals.

$2.7B
Pentesting market (2025)
3.5M
Unfilled cybersecurity positions
75%
Tests still manual
2–6 wk
Typical manual test duration

Manual penetration tests cost between €15K and €80K per audit — prohibitive for SMEs and too infrequent for enterprises under growing compliance mandates like DORA and NIS2. Senior pentesters command salaries above €120K/year, and there simply aren't enough of them.


The Solution

CyberAgents bridges the gap with autonomous, AI-driven pentesting that delivers manual-grade depth at machine speed.

🤖

Autonomous Mode

The system executes all 9 pentesting phases without human intervention. Ideal for recurring audits, compliance validation, and continuous security monitoring. Completes HackTheBox Insane machines in 45–90 minutes.

🎯

Semi-Automatic Mode

The pentester directs strategy while CyberAgents executes. A 10x productivity multiplier that enables junior operators to tackle complex audits previously reserved for senior red teamers.

📋

Professional Reporting

Every engagement produces a detailed penetration test report with findings, CVSS scores, attack chain documentation, and actionable remediation recommendations — ready for compliance review.


Technical Architecture

Built on a multi-agent coordination framework designed for complex, multi-step attack chains.

⚙️

58 Specialized Agents

Purpose-built agents for Nmap, Gobuster, Feroxbuster, Metasploit, Hydra, SQLMap, Nuclei, and more. Each agent encapsulates tool expertise and interpretation logic.

🧠

Blackboard + Coordinator

Intelligent coordination via shared Redis state. The Coordinator Agent orchestrates agent assignments, resolves conflicts, and ensures optimal attack path execution.

🌳

Attack Tree + Backtracking

Monte Carlo Tree Search explores multiple attack paths in parallel. When one path fails, backtracking pivots to alternative vectors — mimicking how expert pentesters think.

Hybrid LLM Routing

Speed-critical tasks route to Gemini Flash for low-latency decisions. Complex reasoning and code generation route to frontier models. Cost-optimized without sacrificing depth.

🔒

Validate-Before-Execute

A validation gate before each Metasploit module or destructive action. Prevents false-positive exploitation and ensures safe, controlled testing within defined scope.

🖥️

Air-Gapped Deployment

Runs entirely on-premise with local LLMs (Qwen, Foundation-Sec-8B) for classified environments, defense, and critical infrastructure. Zero data leaves the perimeter.


9 Pentesting Phases

CyberAgents executes the complete PTES-aligned pentesting methodology end-to-end.

Reconnaissance

Port scanning, service detection, OS fingerprinting, OSINT gathering.

Enumeration

Virtual host discovery, directory brute-force, technology fingerprinting, user enumeration.

Vulnerability Analysis

CVE identification, dependency analysis, template analysis, SSTI/XSS/SQLi testing.

Exploitation

Real exploitation with full proof-of-concept. SSTI → RCE, auth bypass, credential extraction.

Post-Exploitation

Database dumps, credential harvesting, configuration extraction, internal service discovery.

Privilege Escalation

Sudo analysis, SUID binaries, kernel exploits, token abuse, FUSE filesystem tricks.

Lateral Movement

SSH tunneling, port forwarding, credential reuse, internal network pivoting.

Persistence

Backdoor analysis, cron jobs, startup scripts, service implants.

Reporting

Professional PDF report with CVSS-scored findings, attack chain diagrams, and remediation guidance.


Competitive Advantage

CyberAgents is not a vulnerability scanner — it's an AI pentesting team.

CyberAgents Manual Pentest Auto Scanners
Time per test90–360 min2–6 weeks2–4 hours
DepthHigh (9 phases)Very highLow (surface)
Unit cost~€50€15K–80K€200–500
ScalabilityUnlimitedLimited by HRHigh
Real exploitationYes (full PoC)YesNo (detect only)
24/7 coverageYesNoYes (limited)

The Proof: HTB Insane Machines — Hacked Autonomously

No other AI pentesting tool has achieved this. CyberAgents autonomously completed full attack chains on HackTheBox "Insane" machines — including multi-step exploitation, chroot jail escapes, CVE-based privilege escalation, and FUSE filesystem tricks. These are the same challenges that take elite human pentesters days to solve.

💀

Cobblestone (Insane / Linux)

11-step attack chain: Stored XSS → SSTI → RCE as www-data → MySQL credential extraction → hash cracking → chroot jail SSH → internal Cobbler discovery → CVE-2024-47533 auth bypass → Cheetah template injection → root. 4 Critical + 3 High vulnerabilities.

Full Report →
💀

Skyfall (Insane / Linux)

16-step attack chain: Nginx ACL bypass → MinIO metrics exposure → CVE-2023-28432 → S3 bucket enum → Vault token from version history → SSH OTP → sudo vault-unseal → FUSE/sshfs permission bypass → root Vault token → root. 3 Critical + 2 High vulnerabilities.

Full Report →