
Chinese cybersecurity company Qi An Xin is using artificial intelligence to automate vulnerability discovery, penetration testing and security operations as increasingly capable AI models create new challenges for cyber defence.
The company’s approach was profiled by the South China Morning Post on November 20, amid growing reports of AI-enabled hacking incidents affecting banks and government institutions. The report highlighted the growing role of agentic AI in cybersecurity and the company’s efforts to prepare for attacks involving increasingly autonomous systems.
The threat is becoming more pronounced as AI models are used to identify software vulnerabilities that attackers can subsequently exploit. The SCMP report cited the discovery of critical PHP flaws using advanced AI models by an Israeli-American cloud security company late last month, followed by an increase in hacking attempts. PHP is estimated to power approximately 70 per cent of websites.
Qi An Xin’s defensive architecture is built around its proprietary QAX-GPT security large language model, which supports specialised tools for vulnerability analysis, threat investigation and automated security operations. Its AISOC intelligent security operations platform uses the model to analyse alerts, investigate incidents and coordinate responses, reducing the amount of manual work required from security teams.
One of the company’s central developments is Qcode Agents, an AI-powered code security system upgraded in May 2026 to identify vulnerabilities in operating systems, including kernel-level and driver-level software. The system combines source-code and binary analysis with specialised cybersecurity knowledge and dynamic testing to verify whether suspected vulnerabilities can actually be exploited.
Qi An Xin reported that Qcode Agents identified several high-risk vulnerabilities during internal testing, including weaknesses in security components of a domestically developed operating system and Windows kernel drivers. In one case, the system identified a privilege-escalation vulnerability that could allow an ordinary user to obtain root access. The company said it submitted the finding to the affected software vendor for remediation.
The company also released AI Gatling on July 9, an automated penetration-testing agent designed to simulate attacks against computer systems. It uses four specialised AI agents responsible for task planning, attack execution, autonomous reassessment and report generation. The system can also coordinate additional agents for reconnaissance, vulnerability scanning, exploitation and post-exploitation testing.
According to Qi An Xin, AI Gatling can reduce penetration-testing exercises that previously required several days or weeks of human work to a matter of hours. The system is intended to help organisations identify weaknesses before attackers discover and exploit them. The company says it has already been deployed across several sectors, including banking, telecommunications, energy, government and transportation.
AI-driven alert management is another area where the company is reporting progress. During a national AI-enabled cybersecurity application test in September, Qi An Xin said its systems achieved first place in intelligent vulnerability discovery and security alert-log noise reduction. Its proposed alert-management system uses network-side and endpoint-side models to remove duplicate alerts, correlate related events and classify security incidents.
The company reported an alert-noise reduction rate exceeding 99 per cent, an 80 per cent reduction in investigation time and response times of less than five minutes. These figures are based on the company’s account of its testing results. China’s Cyberspace Administration confirmed that Qi An Xin teams achieved strong results in the vulnerability-discovery and alert-log noise-reduction categories, although the government announcement did not publish individual rankings.
Qi An Xin is also developing security controls for organisations adopting autonomous AI agents. Its OpenClaw Security Companion monitors agent activities, controls permissions and helps identify potentially malicious behaviour involving AI tools and plugins. The accompanying SafeSkill platform scans agent skills for security risks that could expose enterprise systems to malicious instructions, data theft or unauthorised operations.
On April 4, the company announced that its OpenClaw Security Companion had passed an assessment conducted by the China Academy of Information and Communications Technology. The evaluation covered nine indicators across infrastructure security, permissions and data security, and application security. Qi An Xin said the product passed all assessment items, including sandbox isolation, permission controls, protection against prompt-injection attacks and plugin security.
Discover more from Aree Blog
Subscribe now to keep reading and get access to the full archive.

