Cyberespionage Campaign Revealed: Chinese Hackers Deploy AI in Sophisticated Operations
A recent report from threat intelligence firm Hunt.io has brought to light a highly sophisticated cyberespionage campaign that is believed to be orchestrated by Chinese-speaking hackers. This alarming revelation indicates that the attackers not only leveraged traditional methods but also integrated commercial AI models into their live attack operations. Their targets included government entities, educational institutions, and industrial systems across several Asian nations, amplifying the concerns about cybersecurity in the region.
Analyzing the data retrieved from five accidentally exposed open directories, researchers at Hunt.io were able to reconstruct the extent and intricate details of the hackers’ operations. Notably, the campaign successfully infiltrated archives belonging to Taiwan’s Kuomintang Party, as well as the Indonesian Ministry of Foreign Affairs, in addition to compromising government and educational systems in mainland China and various industrial hosts in Vietnam. The insights gathered from the open directories revealed the comprehensive orchestration system used by the attackers, shedding light on the advanced tactics employed.
Central to the attackers’ strategy was the deployment of an advanced hacking framework known as SecFlow. This framework allowed the hackers to seamlessly switch between various AI models, namely Claude, Qwen, and DeepSeek, without altering the fundamental task interface. Rather than relying on official APIs for their operations, the attackers opted to channel their requests through private proxy servers linked to the niestools.com domain. This strategic choice meant that the AI models themselves did not autonomously infiltrate systems; instead, they automated various traditional hacking tasks such as vulnerability scanning, credential testing, exploit deployment, webshell installation, data collection, and the generation of attack reports. By effectively linking AI capabilities with live intrusions, the attackers demonstrated a sophisticated understanding of both technology and cybersecurity practices.
The most significant breach reported targeted the Office Automation environment of the Fengtai District government in China. Hackers gained unauthorized access by executing Windows commands through internet-facing web applications. They established a persistent operational backbone by deploying ASPX web shells, which facilitated ongoing access to the compromised systems. To evade detection, the attackers ingeniously divided an LSASS memory dump into 37 segments, extracting vital pieces of information such as password hashes found within the SAM and SYSTEM registry hives. Impressively, they managed to pull 822 user account records and established a new privileged account for backup access. Furthermore, they exfiltrated an alarming 1.28GB of sensitive data, which included a report on chronic diseases that contained real patient health information. In a separate incident, hackers made their way into the management backend of an education-focused AI platform in China, gaining access to 23 AI agent configurations, 14 API secret fields, and 104 chatbot conversations that involved student data.
In executing these attacks, the hackers showcased advanced techniques designed to evade detection. One of their notable strategies involved utilizing a fake MySQL server as bait, intended to trigger unsafe object deserialization in vulnerable Java applications. This method effectively transformed outbound database queries into inbound code execution opportunities. Moreover, they deployed a custom webshell framework referred to as GLUTTON, which cleverly concealed payloads within PNG image files using steganographic techniques. By embedding executable code within the color channel data of these images, the hackers were able to load their code directly into memory without triggering conventional security protocols that scan for malicious files.
This discovery marks the second cyber campaign in just two months in which researchers have identified commercial AI coding tools involved in nation-state intrusions. The previous incident, also uncovered by Hunt.io, showcased the use of Claude Code and DeepSeek targeting entities in Afghanistan and Thailand. In light of these developments, cybersecurity defenders are urged to prioritize the patching of Java frameworks that are vulnerable to deserialization attacks, older Apache or Grafana systems, and public-facing office automation software. As the tactics of cybersecurity attackers evolve, it becomes increasingly essential for organizations to recognize that these adversaries now view commercial AI models as interchangeable tools for operational effectiveness. This shift necessitates a proactive approach, focusing on detecting attack patterns rather than relying solely on identifying specific AI model signatures.
As concerns regarding cybersecurity heighten, the revelations stemming from this investigation reflect a pressing need for enhanced vigilance and security measures across industries. The interplay of AI with traditional hacking techniques presents new challenges for defenders and emphasizes the importance of recognizing and adapting to emerging threats in the ever-evolving landscape of cyber warfare.