Russian-language cybercrime group CyberXero has deployed AI tools, including Anthropic’s Claude Code, to automate and scale cyberattacks. The operation resulted in the theft of over 628,000 personal records from Ukrainian citizens and attacks on critical infrastructure, including a district heating plant in Kharkiv.
Russian-language cybercrime group CyberXero has deployed AI tools, including Anthropic’s Claude Code, to automate and scale cyberattacks. The operation resulted in the theft of over 628,000 personal records from Ukrainian citizens and attacks on critical infrastructure, including a district heating plant in Kharkiv.
This is stated in a report by cybersecurity researchers from SOCRada.
According to SOCRadar analysts, the attackers, who have been operating as Initial Access Brokers since at least July 2026, have been implementing two lines of attack in parallel, The Hacker News writes .
In total, data theft from four Ukrainian organizations has been confirmed, resulting in the leakage of over 628,000 personal records. Compromise of resources in Poland, China, and the Netherlands has also been recorded.
The attack is characterized by a high level of integration of artificial intelligence into the cyber operations combat cycle. The hackers used a two-tier AI ecosystem:
a cluster of 51 specialized agents based on Claude Code for automatic scanning, exploit chain construction, and reconnaissance;
a standalone PentAGI system connected to the Cobalt Strike command server via an API for automatic code generation and execution by the attack vector.
To circumvent the built-in security limitations of AI models, attackers used context engineering and regular session resets. They convinced the system that they were performing “authorized penetration testing of their own systems.” If the model refused to perform the malicious task, the hackers simply opened a new session with a pre-prepared “safe” context.
However, SOCRadar noted that AI security tools worked successfully in several cases — the model refused to install a backdoor, disable a firewall, or perform lateral movement through the network four times in a row.
The operation was exposed after researchers found an unsecured directory on the attackers’ server (46.21.250[.]135). It contained AI session logs, source code for scripts with exposed tokens, exploitation tools, and stolen victim data.
In addition, SOCRadar reported on another large-scale cybercrime operation, Operation Master (hacker nicknamed masterblack). In this campaign, attackers used vulnerabilities in VPN gateways (including CVE-2026-0257 in Palo Alto Networks GlobalProtect) and AI tools to develop phishing platforms and account spoofing. Companies and citizens in Latin America and Japan became victims of financial fraud worth millions of dollars.
