AI-Powered Attacks Hit FortiGate VPNs Using DeepSeek and Claude Automation


Threat actors deployed DeepSeek and Claude LLMs to breach 2,500+ FortiGate SSL VPN devices across 106 countries. Custom ARXON MCP server and CHECKER2 Docker orchestrator automate mass scanning, vuln assessment, and exploitation. Misconfigured attack server exposed full pipeline in early February 2026.

Stolen FortiGate credentials fed into DeepSeek generate strategic attack plans. Claude executes coding tasks running Impacket and Metasploit autonomously post-access. Low-skill operators manage thousands of targets simultaneously.

Exposed server revealed parallel batch processing logs. Telecom, finance, and government edge devices hit worldwide. Automation scales beyond human capacity dramatically.

CHECKER2 scans VPN endpoints at scale through Docker containers. ARXON bridges network recon data into LLM prompts producing actionable exploits. Vulnerability reports auto-prioritize privilege escalation paths.

Traditional defenses fail against AI-orchestrated speed. Manual SOC response cannot match parallel automation across continents.

Attack Infrastructure Table

ComponentFunctionLLM Role
CHECKER2Docker VPN scannerTarget discovery
ARXONMCP serverData-to-exploit bridge
DeepSeekStrategy generationRecon → attack plans
ClaudeCode executionImpacket/Metasploit runs

Telecom perimeters breached systematically. Exposed deploy_output.log showed global target diversity. Single operator controls thousands via AI pipeline.

Intrusion chain (Source – Cyber and Ramen)

Kill Chain Automation

Recon phase: CHECKER2 maps FortiGate endpoints globally
Planning phase: DeepSeek analyzes stolen configs for attack paths
Execution phase: Claude writes/runs Impacket, Metasploit exploits
Privilege phase: Auto-generated escalation recommendations
Lateral phase: Network mapping for critical asset discovery

Redacted snippet of the vulnerability assessment report found on the server (Source – Cyber and Ramen)

Custom MCP protocol feeds structured network data into LLMs. Output formats exploit chains directly without human parsing.

Exposed vuln assessment snippets show model reasoning. Prioritized next steps include Kerberoasting, LSASS dumping, unquoted service paths.

Snippet of the contents of deploy_output.log showing thousands of targets across the world (Source – Cyber and Ramen)

Immediate Defense Actions

  • Patch all FortiGate firmware to latest stable immediately
  • Audit VPN accounts for anomalous creations weekly
  • Monitor SSH sessions from unexpected sources
  • Deploy network baselines against config drift
  • Block ARXON MCP traffic patterns at perimeter
  • Hunt Impacket execution through EDR logging

AI kill chain represents cybercrime evolution. Human oversight drops to strategic direction only. Scale defeats traditional capacity-based defenses completely.

Exfiltration targets crown jewels post-access. Credential harvesting precedes ransomware or data theft systematically.

FAQ

What devices did attackers target?

FortiGate SSL VPN appliances using stolen configuration data.

Which AI models powered the attacks?

DeepSeek for planning, Claude for code execution and vuln assessment.

How many targets processed simultaneously?

2,500+ devices across 106 countries in parallel batches.

What custom tools drove automation?

CHECKER2 Docker scanner + ARXON Model Context Protocol server.

Which sectors faced highest risk?

Telecom, finance, government edge infrastructure worldwide.

How to detect these AI attacks?

Audit VPN accounts, monitor SSH, baseline configs, hunt Impacket.

Readers help support VPNCentral. We may get a commission if you buy through our links. Tooltip Icon

Read our disclosure page to find out how can you help VPNCentral sustain the editorial team Read more

User forum

0 messages