SOCRadar Details CyberXero’s AI-Assisted Intrusions and Web Exploitation

· Original article ↗

Summary

SOCRadar researchers detail CyberXero, an active Russian-speaking initial access broker using mass web-app exploitation and targeted intrusions, including against Ukrainian utilities. The profile documents its AI tooling, techniques, victim impact and defenses.

Key points

  • SOCRadar says it identified more than 40 affected organizations across several countries, excluding more than 4,500 WordPress sites targeted through mass exploitation.
  • The in-house wp2shell tool exploits a WordPress REST API batch-endpoint authorization desynchronization to use SQL injection, create administrator accounts and deploy a webshell.
  • The report documents targeted intrusions involving Ukrainian organizations, including alleged theft of subscriber records and credentials; it also describes reconnaissance against Ukrainian energy and utilities entities.
  • Researchers found a setup involving 51 Claude Code agents and PentAGI connected to a Cobalt Strike Team Server through the Anthropic API.
  • In a Chinese corporate network, the report describes Redis master-replication abuse to plant an SSH key, followed by lateral movement; successful EternalBlue exploitation was not confirmed.
  • The article recommends checking for wp2shell artifacts, patching exposed applications, removing default credentials, securing Redis, and protecting AI-agent session logs.

Article Details

Attack Vectors
  • Mass exploitation of WordPress REST API batch endpoint authorization desynchronization using adaptive SQL injection, followed by creation of rogue administrator accounts and deployment of WSO-family webshells.
  • Targeted intrusions used blind SQL injection, password-reset token extraction, and PHP deserialization to obtain remote code execution and dump databases.
  • Compromised exposed Magento configuration and used hardcoded administrator credentials to access victim systems and data.
  • Abused a misconfigured Redis server with rogue master replication to plant an SSH key, then moved laterally using WMI through a passwordless account.
  • Attempted password spraying against Pakistani government entities and brute-forced a ZTE gateway using default credentials.
  • Used a card-skimmer endpoint against a production payment flow and delivered Cobalt Strike beacons, staging, and PowerShell stagers from the Team Server.
Defensive Notes
  • Review WordPress user tables for usernames matching wp2_ followed by eight hexadecimal characters, and plugin paths matching wp2shell_ with the same suffix; treat either artifact as evidence of compromise.
  • Prioritize patching CVE-2026-63030, CVE-2026-60137, and CVE-2026-4815; verify that Magento configuration files such as app/etc/env.php are not publicly reachable.
  • Audit administration panels for hardcoded or shared credentials, replace default network-device passwords, and remove local accounts with blank passwords.
  • Do not expose Redis to the internet; require authentication, disable or rename SLAVEOF/REPLICAOF where unnecessary, and monitor authorized_keys for unexpected changes.
  • Block the listed C2 infrastructure and search historical logs for connections to 42.193.227.214 on ports 1002, 8044, and 8033; hunt for the UpdSvc service and PowerShell download cradles.
  • Protect AI-agent session logs with retention, encryption, and access controls comparable to those used for credential stores.

Indicators of compromise

TypeIndicatorContext
IPV4212[.]193[.]31[.]189Attacker-operated cryptocurrency payment gateway.
IPV442[.]193[.]100[.]94Attacker Redis-exploitation node and PentAGI agent infrastructure.
IPV442[.]193[.]227[.]214Cobalt Strike Team Server and PentAGI infrastructure; receives HTTPS beacons on port 1002, serves HTTP staging on port 8044, and delivers PowerShell stagers on port 8033.
IPV445[.]88[.]106[.]2Attacker provisioning server, reported on port 1500.
IPV445[.]88[.]106[.]78Attacker mass-scanning server.
IPV446[.]21[.]250[.]135Attacker-controlled primary workstation and open-directory seed node.
IPV449[.]234[.]12[.]182Attacker staging and payload-delivery server used in the Chinese network cluster.
IPV491[.]208[.]184[.]148Attacker-controlled secondary workstation.
MD53870a559b5b6ea6076f5f90029c180b7Hash of the Cobalt Strike x86 stager32.bin.
SHA25659c535f47ab4d35f6d9fc8b8aec4a72438ea0b89e5a4dcea74b05d2d32cfa483Hash of config.bin, identified as a Chinese-cluster staging payload.
SHA25671e21094c1ac1cf0275c91ed377965e248bca1c12711f6dd20e2682681fc1192Hash of 1.bin, identified as a Chinese-cluster staging payload.
SHA25692a789444708fa1cb4cc5a89e0aa6cc7a279b62a7b8a4d2857255a18197d0090Hash of the Cobalt Strike x86 stager32.bin from the pirated 4.9.1 build.
SHA256a5ae0aab352871bc0b038b3aa43b03eb223425628c884af7b9fc592cd34eb86cHash of an obfuscated WSO webshell deployed by the actor.

MITRE ATT&CK

T1046 · Network Service DiscoveryAfter gaining access to the Chinese corporate network, beacons enumerated the network.T1047 · Windows Management InstrumentationThe actor moved laterally through a Chinese corporate network over WMI using a passwordless account.T1056.003 · Web Portal CaptureThe actor built and tested a card-skimmer endpoint against a production payment flow.T1059.001 · PowerShellThe Cobalt Strike infrastructure delivered PowerShell stagers.T1071.001 · Web ProtocolsThe Cobalt Strike Team Server received HTTPS beacons on port 1002.T1098.004 · SSH Authorized KeysThe actor abused Redis master replication to plant an SSH key.T1110.003 · Password SprayingThe actor attempted credential spraying against two Pakistani government entities.T1190 · Exploit Public-Facing ApplicationThe actor exploited internet-facing WordPress applications through SQL injection and abused vulnerabilities in other exposed web applications.T1505.003 · Web ShellAfter compromising WordPress sites, the actor deployed a WSO-family webshell registered as an active plugin.T1543.003 · Windows ServiceThe actor created the UpdSvc Windows service for persistence in the Chinese network cluster.T1588.005 · ExploitsThe actor had functional exploits for disclosed vulnerabilities, including an exploit for CVE-2026-4815 developed within 30 days of publication.T1595.002 · Vulnerability ScanningThe actor scanned more than 56,000 URLs and ran an automated scan against 4,708 targets to find exploitable WordPress panels.

CVE

Threat Actors

Malware

Vendors

Products

Tools

Countries

Industries

Related Articles