Chinese‑language hackers deploy Claude, Qwen and DeepSeek AI agents in new Asian intrusion campaign

Security researchers have identified a fresh wave of cyber‑espionage activity in which Chinese‑speaking threat actors are integrating large‑language‑model based AI agents—specifically those powered by Claude, Qwen and DeepSeek—into their intrusion toolkits.

The AI agents appear to be used as autonomous assistants that can perform reconnaissance, generate phishing content, and even manipulate compromised systems without direct human input. By leveraging the natural‑language capabilities of these models, the operators can script complex attack sequences more rapidly than traditional manual methods.

Targets span a broad spectrum of Asian institutions, including government ministries, political organizations, universities and a variety of industrial enterprises. The selection of sectors suggests an interest in gathering strategic intelligence, influencing policy discussions and potentially exfiltrating proprietary technology.

Analysts note that this operation differs from an earlier campaign that also employed AI tools, indicating a deliberate evolution in tactics. While the prior effort relied on more generic automation scripts, the current intrusion chain embeds conversational agents that can adapt to defensive measures in real time.

The emergence of AI‑augmented malware raises new challenges for defenders. Traditional detection signatures may miss the dynamic code generated on the fly by language models, and the ability of agents to craft context‑aware social‑engineering messages could increase the success rate of phishing attempts.

Cyber‑security firms are urging organizations to adopt stricter monitoring of outbound AI service calls, enforce network segmentation, and train staff to recognize AI‑enhanced phishing cues. Enhanced threat‑intel sharing across the region is also being recommended to track the evolving use of generative AI in hostile operations.

As large‑language models become more accessible, experts warn that the line between automated scripting and fully autonomous hacking will continue to blur, prompting a reassessment of defensive postures worldwide.