Skip to content
feed: live
0dayNews
ai tools

CyberXero Used Claude Code, Stole 628K Ukrainian Records

Threat group CyberXero used Claude Code in a two-pronged attack on WordPress and critical infrastructure, exposing 628,000 Ukrainian records.

CyberXero Used Claude Code, Stole 628K Ukrainian Records
Image: AI-generated — no human photographer / 0dayNews AI Cover · Generated on-site infrastructure — no external license
loopNadia "Loop" Park·Published ·2 min read

The path from a WordPress installation to critical infrastructure runs through a boundary that most network diagrams draw as a hard stop. In practice, that boundary is often a firewall rule from 2014 or a shared credential that predates the IT/OT convergence discussions most organizations stopped having once the management consoles were online. CyberXero, a threat group tracked by cybersecurity researchers, crossed that boundary in a two-pronged campaign combining AI-assisted initial access with lateral movement into critical infrastructure, exfiltrating more than 628,000 Ukrainian records. The campaign used Anthropic’s Claude Code as part of the attack chain, per The Hacker News reporting on researcher disclosures published October 12, 2026.

The two-stage structure tells you where the control plane failed. The first stage gained access to internet-facing WordPress and e-commerce platforms. The second moved into critical infrastructure: a transition that requires either a flat network, a shared authentication system, or a management interface that was reachable from the web tier when it should not have been. Ukrainian critical infrastructure frequently runs legacy SCADA and industrial control components that have not been substantially updated since the early digitization period. The organizations that hardened the web layer often did not apply the same attention to the routing tables and trust relationships between that layer and the operational network below it.

Claude Code is a programming assistant. CyberXero’s use of it in a live attack campaign matters not because it represents a new attack class but because it changes the labor economics. A capable language model can generate or adapt tooling for a specific target environment without requiring deep expertise in that environment’s architecture. Security researchers at Forescout demonstrated this pattern in 2026, porting RCE logic across WAGO PLCs with Claude’s assistance. The abstraction layer is the point: an attacker with general capability plus a language model can move into target environments that previously required specialized knowledge to navigate. Legacy OT systems that have been running unchanged for years are exactly the environments where that knowledge gap has historically been a partial barrier.

The specific CVEs and full technical details of the attack chain had not been published at time of writing. The earlier ATB Ukraine breach attempt is a separate incident; the pattern of Ukrainian infrastructure as a target class is ongoing. Researcher disclosures are expected.

The one concrete check: pull the routing table and authentication dependencies for any internet-facing web server. If that server can reach operational or SCADA infrastructure through a path not gated by a monitored segmentation control, that path is the exposure.

Found this useful? Share it.