A Chinese-speaking hacker has been identified using an AI-powered hacking tool that inadvertently exposed its full infrastructure, offering a rare glimpse into how attackers are incorporating large language models (LLMs) in real-world campaigns This article explores attacker ai agent. . Researchers from Unit 42 discovered this activity after the attacker’s AI agent mistakenly launched a public file server from its working directory, revealing sensitive assets including exploit scripts, API keys, configuration files, and attack logs.
The threat actor, operating under aliases “knaithе” and “KnYuan,” used DeepSeek integrated with the open-source Hermes Agent framework as the core autonomous operator. The system identified PoC exploits that matched exposed services, leading it to choose n8n workflow automation software, which was chained with CVSS 9.9.
Despite these failures, the workflow showcases a fully autonomous attack loop capable of reconnaissance, decision-making, and exploitation attempts without continuous human input. The campaign's most significant turning point occurred when the Hermes Agent initiated a local HTTP server using: In addition to DeepSeek, the actor also explored several LLMs including Qwen, GLM, Kimi, and MiniMax. The ability to swiftly identify, prioritize, and attempt exploitation across extensive attack surfaces significantly shortens the time and effort required for cyberattacks.
By cutting through SOC investigation blind spots and containing threats earlier, ANY.RUN can help reduce response costs and business disruptions.
