Amazon: AI-Assisted Attacker Breached 600 FortiGate Firewalls in Five Weeks

Summary
Amazon says a Russian-speaking actor used AI-assisted tools and weak credentials to breach more than 600 exposed FortiGate firewalls in 55 countries over five weeks; researchers detail reconnaissance tools and targeting of Veeam backups.
Key points
- Amazon reported that the campaign ran from January 11 to February 18, 2026, and affected more than 600 FortiGate firewalls across 55 countries.
- The actor targeted internet-exposed management interfaces and used common-password brute force; Amazon says no exploits were used to breach the firewalls.
- Stolen firewall configurations contained VPN and administrative credentials and network information that could support access to victim networks.
- Researchers identified ARXON, a custom MCP server, and CHECKER2, a scanning orchestrator, as tools used to connect reconnaissance data with LLM-generated attack plans.
- The campaign targeted Veeam backup servers with custom tools and attempted to exploit known vulnerabilities; the article does not establish that those attempts succeeded.
- A separate researcher reported finding a server with 1,402 files, including FortiGate configurations, credential dumps, and attack-planning documents.
- Amazon recommends keeping management interfaces off the internet, enabling MFA, avoiding reused VPN and Active Directory passwords, and hardening backups.
Article Details
- Attack Vectors
- The actor scanned internet-exposed FortiGate management interfaces on ports 443, 8443, 10443, and 4443, then used brute-force attacks with common passwords against devices lacking MFA.
- After gaining access, the actor extracted FortiGate configurations containing VPN and administrative credentials, network topology, routing information, and firewall policies, then used VPN access to enter victim networks.
- Custom Go and Python tools automated reconnaissance, including routing-table analysis, network classification, port scanning, and identification of SMB hosts and domain controllers.
- The actor targeted Veeam Backup & Replication servers with custom PowerShell scripts and credential-extraction tools, and reportedly attempted exploits for CVE-2019-7192, CVE-2023-27532, and CVE-2024-40711.
- Reconnaissance data was fed through the custom ARXON MCP server to LLMs, including DeepSeek and Claude, for attack planning and lateral-movement guidance.
- Defensive Notes
- Do not expose FortiGate management interfaces to the internet; enable MFA and avoid reusing VPN passwords for Active Directory accounts.
- Harden backup infrastructure and keep edge devices patched.
- Audit unusual SSH activity and VPN account creation.
Indicators of compromise
| Type | Indicator | Context |
|---|---|---|
| IPV4 | 212[.]11[.]64[.]250 | Server attributed in the report to the threat actor, hosting tools and a PowerShell script used to target FortiGate and Veeam systems; the exposed server also contained stolen configurations and other intrusion data. |
MITRE ATT&CK
T1003.006 · DCSyncOperational documentation described using Meterpreter and mimikatz for DCSync attacks to extract NTLM hashes from Active Directory; the article does not confirm that these steps succeeded.T1016 · System Network Configuration DiscoveryThe reconnaissance tools analyzed routing tables and network topology on compromised networks.T1018 · Remote System DiscoveryThe tools identified hosts, including SMB hosts and domain controllers, on victim networks.T1046 · Network Service DiscoveryThe actor scanned VPN targets and used port scans during reconnaissance of compromised networks.T1059.001 · PowerShellCustom PowerShell scripts were used to target Veeam Backup & Replication servers.T1110.001 · Password GuessingThe actor used brute-force attacks with common passwords against exposed FortiGate devices.T1133 · External Remote ServicesThe actor targeted internet-exposed management interfaces and used VPN access to enter victim networks.T1190 · Exploit Public-Facing ApplicationThe actor reportedly attempted to exploit Veeam vulnerabilities, including CVE-2023-27532 and CVE-2024-40711.T1552.001 · Credentials In FilesThe actor extracted and parsed FortiGate configuration files containing recoverable VPN and administrative credentials.
CVE
CVE-2019-7192Qnap · Photo stationnotes” also contained multiple references to trying to exploit various vulnerabilities, including CVE-2019-7192 (QNAP RCE), CVE-2023-27532 (Veeam information disclosure), and CVE-2024-40711 (Veeam RCE).CVE-2023-27532Veeam · Veeam backup & replicationmultiple references to trying to exploit various vulnerabilities, including CVE-2019-7192 (QNAP RCE), CVE-2023-27532 (Veeam information disclosure), and CVE-2024-40711 (Veeam RCE).CVE-2024-40711veeam · backup_\&_replication; Veeam · Backup and Recoveryincluding CVE-2019-7192 (QNAP RCE), CVE-2023-27532 (Veeam information disclosure), and CVE-2024-40711 (Veeam RCE).
People
CJ MosesCISO of Amazon Integrated Security and the source of the report describing the campaign and its findings.German FernandezCronUp security researcher who found a separate server exposing what appeared to be AI-generated tools targeting FortiWeb; the article says those tools did not appear to be part of the FortiGate campaign.
Vendors
FortinetA new report by CJ Moses, CISO of Amazon Integrated Security, says that the hacking campaign occurred between January 11 and February 18, 2026, and did not rely on any exploits to breach Fortinet firewalls.Veeamand LLMs) to automate reconnaissance, lateral movement, and attacks against backup infrastructure like Veeam, prompting recommendations to close exposed interfaces, enable MFA, and harden backups.
Products
ClaudeTwo folders named claude-0 and claude contained over 200 files between the two, including Claude Code task outputs, session diffs, and cached prompt states.”Claude CodeTwo folders named claude-0 and claude contained over 200 files between the two, including Claude Code task outputs, session diffs, and cached prompt states.”DeepSeekappliances and internal networks was fed into ARXON, which then queried large language models such as DeepSeek and Claude to generate structured attack plans.Fortinet FortiGateA Russian-speaking threat actor used multiple generative AI services to breach more than 600 FortiGate firewalls across 55 countries in five weeks by targeting exposed management interfaces and weak credentials ratherMicrosoft Active DirectoryOperational documentation written in Russian detailed how to use Meterpreter and mimikatz to conduct DCSync attacks against Windows domain controllers and extract NTLM password hashes from the Active Directory database.Veeam Backup & ReplicationThe campaign also specifically targeted Veeam Backup & Replication servers using custom PowerShell scripts, compiled credential-extraction tools, and attempted to exploit Veeam vulnerabilities.
Tools
ARXONA custom MCP server (ARXON) and tools like CHECKER2 fed reconnaissance into LLMs to automate large-scale scanning and attack planning, though AI-generated code showed reliability issues.CHECKER2A custom MCP server (ARXON) and tools like CHECKER2 fed reconnaissance into LLMs to automate large-scale scanning and attack planning, though AI-generated code showed reliability issues.gogo scannerby analyzing routing tables, classifying networks by size, running port scans using the open-source gogo scanner, identifying SMB hosts and domain controllers, and using Nuclei to look for HTTP services.hashcatIn some cases, Claude Code was configured to execute offensive tools on its own, including Impacket scripts, Metasploit modules, and hashcat, without requiring the threat actor to approve each command.HexStrike MCPThe researcher notes that the operation evolved over several weeks, with the attacker initially using an open-source HexStrike MCP framework and, about eight weeks later, moving to the automated, customized ARXON system.ImpacketIn some cases, Claude Code was configured to execute offensive tools on its own, including Impacket scripts, Metasploit modules, and hashcat, without requiring the threat actor to approve each command.Metasploit FrameworkIn some cases, Claude Code was configured to execute offensive tools on its own, including Impacket scripts, Metasploit modules, and hashcat, without requiring the threat actor to approve each command.MeterpreterOperational documentation written in Russian detailed how to use Meterpreter and mimikatz to conduct DCSync attacks against Windows domain controllers and extract NTLM password hashes from the Active Directory database.MimikatzOperational documentation written in Russian detailed how to use Meterpreter and mimikatz to conduct DCSync attacks against Windows domain controllers and extract NTLM password hashes from the Active Directory database.Nucleiport scans using the open-source gogo scanner, identifying SMB hosts and domain controllers, and using Nuclei to look for HTTP services.
