Chinese Hacker Uses DeepSeek for Autonomous Attacks

Palo Alto Networks' Unit 42 reported on March 13, 2024, that a Chinese-speaking threat actor, identified by the aliases knaithe and KnYuan, utilized the large language model DeepSeek to launch autonomous cyberattacks. This operation involved the open-source Hermes Agent framework, which facilitated the independent execution of malicious activities. The attack chain began with an initial command sent via the Telegram messaging application. Following this instruction, the Hermes Agent autonomously identified internet-facing systems that were vulnerable to attack. The agent then proceeded to select and deploy publicly available exploits against these identified targets without requiring further human intervention. Researchers from Unit 42 observed that no additional operator input was recorded during the observed session, indicating a high degree of automation. This marks a significant development in the use of AI models by threat actors, enabling more sophisticated and potentially widespread attacks with reduced direct oversight. The Hermes Agent framework itself is designed to enable AI agents to perform tasks, and its integration with powerful LLMs like DeepSeek amplifies its offensive capabilities. DeepSeek, developed by DeepSeek AI, is a family of large language models known for their strong performance in various natural language processing tasks, including code generation and reasoning, making them suitable for complex operational tasks like exploit selection. The autonomous nature of these attacks poses a considerable challenge to cybersecurity defenses, as they can operate at machine speed and adapt to network conditions without human delay. The reliance on Telegram for initial command and control also highlights the evolving tactics of threat actors in leveraging readily available and encrypted communication platforms. The researchers' inability to detect further operator input suggests that the AI agent was fully responsible for the attack's progression from reconnaissance to exploitation. This incident underscores the growing concern within the cybersecurity community regarding the weaponization of AI technologies by malicious actors and the potential for AI-driven autonomous cyber warfare. The specific exploits used in this attack were not detailed, but the ability of the agent to select them implies a sophisticated understanding of system vulnerabilities and exploit efficacy. The implications of such autonomous attacks extend to a potential increase in the speed and scale of cyber threats, making rapid detection and response even more critical.