A cybercriminal group known as CyberXero reportedly used artificial intelligence (AI) tools alongside traditional hacking software in a campaign targeting websites and critical infrastructure. According to cybersecurity researchers, the operation affected WordPress websites, e-commerce platforms and Ukrainian organizations. More than 628,000 records linked to Ukrainian individuals were reportedly stolen. Researchers believe the activity began at least in July 2026.

The researchers identified CyberXero as a Russian-speaking, financially motivated initial access broker, a type of cybercriminal that gains entry into computer systems and may sell that access to others. The operation followed two separate approaches: automated attacks against websites worldwide and carefully selected attacks against Ukrainian energy and critical infrastructure. The global campaign focused on exploiting known security weaknesses in WordPress websites and e-commerce platforms. Researchers also identified compromises in Poland, China and the Netherlands.

According to the investigation, one automated operation scanned 4,708 targets and gained access to 429 WordPress administration panels. The attackers then deployed 32 web shells within just 61 seconds. Web shells are malicious files that can give attackers remote access to compromised servers. The campaign also involved the exploitation of exposed Redis servers, which resulted in breaches affecting two Chinese organizations.

The AI component played an important role in coordinating the operation. CyberXero reportedly used Anthropic’s Claude Code with 51 specialized AI agents to support scanning, reconnaissance, exploitation and activities after gaining access to systems. The attackers also connected an AI-powered penetration-testing tool called PentAGI with Cobalt Strike infrastructure. Investigators found that the attackers sometimes changed their instructions or started new sessions when the AI refused certain requests.

The investigation also revealed that AI safety controls did not always cooperate with the attackers’ instructions, as several requests were refused. The AI models reportedly declined attempts to install backdoors, disable firewalls, move through connected systems and deploy web shells. Researchers discovered the campaign after finding an exposed directory containing working files, AI session logs, scripts, configuration details and stolen data. This accidental exposure helped investigators understand how the operation was organized.

The incident highlights the growing security risks associated with combining AI agents and traditional hacking tools. Organizations using WordPress, Magento or PrestaShop should review their security updates and check for unfamiliar administrator accounts or suspicious files. Businesses should also protect internet-facing Redis servers and monitor their systems for unexpected access. The investigation shows how AI can help attackers coordinate operations at scale, making strong security controls, regular patching and continuous monitoring increasingly important.

Stay alert, and keep your security measures updated!

Source: Follow cybersecurity88 on X and LinkedIn for the latest cybersecurity news