Generative AI, Retrieval-Augmented Generation (RAG) and Agentic AI in Offensive Cyber Operations

As artificial intelligence (AI) continues to advance, its implications for cybersecurity have become a critical area of concern. The utilization of GenAI/LLMs in support of offensive cyber operations could significantly enhance the scale and effectiveness of cyber threats, particularly when leveraged by state-sponsored actors. Among these actors, the People’s Republic of China (PRC) stands out for its extensive cyber capabilities and strategic use of AI-driven tools in cyber warfare. Understanding how China incorporates GenAI/LLMs into its cyber operations is essential for assessing the evolving threat landscape and the broader geopolitical ramifications. Understanding China’s potential use of GenAI/LLMs for offensive cyber operations is a top priority since China is the most ‘prolific’ adversary in cyberspace, amongst the Big Four, i.e. Russia, China, North Korea, Iran. China has orchestrated the largest volume of cyberattacks against the US in the last few years with CrowdStrike Intelligence observing a 150% increase in the number of China-nexus intrusions in 2024 as compared to 2023. The US Cybersecurity Infrastructure Security Agency (CISA) has “long been concerned about the breadth, depth, and sophistication of the People’s Republic of China (PRC) state-sponsored cyber threat to the United States1”. FBI Director Christopher Wray argues2 that “China’s hacking teams outnumber the FBI’s cyber agents 50 to 13”, and he has labeled China as the “source of more cyber attacks on the US than all other nations combined” with “a lot of funding and sophisticated tools, and often joining forces with cyber criminals – in effect, cyber mercenaries4”. The degree of ‘innovation’ granted by the use of GenAI is unclear, which poses the question of whether LLMs are tipping the pendulum towards granting adversaries novel tools, techniques and practices (TTPs)—and greater strategic advantage–in their offensive cyber operations, or whether these tools are merely multiplier effects, used to enhance, amplify, automate and accelerate the current TTPs at hand, instead of creating new ones. In our first paper, we explored the application of GenAI/LLMs for offensive cyber operations and asserted the real advantage of GenAI/LLMs, at least in its current state, lies not in the creation of novel offensive techniques but in the enhancement of existing ones. Adversaries are able to wield GenAI/LLMs to expand their repertoire of procedures and enhance the efficacy of their techniques, in turn achieving greater operational effectiveness during their operations. This paper is the second in our series that attempts to offer a grounded hypothesis of how an actual adversary could leverage GenAI/LLMs in an intrusion. With China as a principal national security threat, we decided to develop our hypothesis around a prolific China-nexus adversary, namely Aquatic Panda. Based on the known tactics, techniques, and procedures (TTPs) of the adversary, we will elucidate how GenAI/LLMs can be used by Aquatic Panda to enhance their offensive capabilities.