CrowdStrike Exposes China-Linked Hackers Using AI Agents to Target Over 200 Organisations

Cybersecurity company CrowdStrike has revealed a large-scale cyberespionage campaign in which a China-linked threat group allegedly used artificial intelligence agents to attack more than 200 organisations worldwide. In a report published on October 29, 2026, the company identified the group as VAPOR PANDA and said it used an AI-powered framework known as ARTEX to coordinate multiple cyber operations simultaneously. The campaign reportedly began on October 17 and targeted government institutions, foreign ministries, telecommunications providers, pharmaceutical companies, semiconductor businesses and defence-related organisations. CrowdStrike said its investigation uncovered evidence of activity across at least 14 countries during an 11-day period. The findings highlight how cybercriminal groups and state-linked operators may increasingly use AI agents to expand the scale and speed of their operations.

According to CrowdStrike, the attackers primarily relied on China’s DeepSeek large language model alongside another subscription-based commercial AI model that investigators could not identify. The ARTEX framework coordinated several AI agents, including planning and execution agents, allowing them to investigate targets, identify vulnerabilities and pursue different attack routes. Rather than manually controlling every operation, a human operator provided instructions while the agents carried out multiple technical activities in parallel. CrowdStrike reported that one intrusion involved approximately 1.4 billion input tokens consumed by the planning agents, demonstrating the substantial scale of automated activity. The company described this development as an escalation in the use of AI for cyber operations because the technology allowed a relatively small number of operators to manage attacks against numerous organisations at the same time.

The targeted organisations reportedly included foreign ministries in Latin America, Europe, Asia and the Middle East, as well as 19 US-based organisations operating in pharmaceutical, semiconductor, defence, nuclear and financial sectors. CrowdStrike said it confirmed successful post-exploitation activity at nine organisations, where the attackers deployed a backdoor known as BLOODALCHEMY to maintain access to compromised systems. The group also reportedly conducted credential scanning, password attacks and network reconnaissance against additional targets. In some operations, the attackers attempted to obtain administrator credentials, access cloud infrastructure and move between connected systems. CrowdStrike further reported that the AI agents were instructed to identify computers without the company’s Falcon security software, suggesting that the attackers incorporated security-detection avoidance into their automated workflow.

One particularly concerning aspect of the investigation was the evidence that human operators remained involved in directing the AI agents. Recovered transcripts reportedly contained instructions written in Chinese, showing that the operator guided the system towards specific targets and objectives. The agents could then investigate different attack paths and switch approaches when an initial method failed. This combination of human supervision and automated execution raises concerns about how quickly sophisticated cyberattacks could develop. Traditional attacks often require considerable technical expertise and time to investigate multiple networks individually. AI-assisted frameworks may reduce some of that workload, allowing attackers to conduct reconnaissance, test possible weaknesses and coordinate operations at a much greater scale. However, CrowdStrike’s findings do not mean that the AI independently carried out the entire campaign; human operators continued to provide direction and make important decisions.

CrowdStrike warned that China-linked threat actors are likely to continue experimenting with agentic AI frameworks over the next 12 months, potentially expanding their ability to conduct simultaneous cyber intrusions. The company recommended that organisations strengthen their security systems by regularly updating software, rotating exposed credentials, protecting cloud infrastructure and ensuring that endpoint detection systems cover all devices and servers. The report also reinforces the importance of monitoring AI tools that can access networks, execute commands and interact with external systems. Although AI agents offer significant benefits for legitimate businesses and cybersecurity professionals, their misuse could create new challenges for governments, financial institutions and technology companies. The latest campaign demonstrates that the cybersecurity threat landscape is changing as artificial intelligence becomes increasingly integrated into both defensive and offensive operations.