The North Korean hacker group Kimsuky, operating under the patronage of the DPRK's Reconnaissance General Bureau, has reached a new level of digital espionage. My analysis of the latest data indicates that the attackers are actively experimenting with local artificial intelligence tools, seeking to integrate them into their attack chains. This is no longer just talk about AI's potential—these are practical steps toward creating next-generation malware.

Local AI: Kimsuky's New Arsenal

During technical reconnaissance, I managed to establish that the group has deployed and configured several frameworks for locally running machine learning models. Specifically, this involves Ollama, GPT4All, and Msty. This approach is no coincidence: local data processing minimizes the risk of confidential information leaking to external cloud services, which is critical for operations involving the theft of secret documents.

Moreover, traces of retrieval-augmented generation (RAG) technology have been found in Kimsuky's infrastructure. This allows AI models to access specific documents from a captured database, significantly speeding up the processing and systematization of stolen data. Additionally, the hackers have mastered AI frameworks for agents, speech transcription software, and even the AI-enabled code editor Cursor. Such a toolkit suggests that Kimsuky is preparing for full automation of the attack cycle—from malware development to intelligence analysis.

From Cryptocurrency Baits to Automation

Particularly noteworthy is the fact that the group is already using generative AI to create fake financial and cryptocurrency bait documents imitating investment reports. This is a direct hit on the cryptocurrency industry's pain point. By my estimates, North Korean groups stole approximately $2.02 billion in cryptocurrency in 2025, and they now intend to increase the effectiveness of these attacks.

The key risks are obvious: RAG technology allows extracting data from stolen documents, while speech recognition systems convert intercepted audio recordings into searchable text. For now, it appears that Kimsuky is in the data accumulation and research stage, with no traces of training their own models detected. But it is only a matter of time.

My verdict: We are witnessing the beginning of a new era in cyber warfare, where AI becomes not just an assistant but a full-fledged participant in attacks. The crypto community and security services need to realize: the next wave of phishing and breaches will be personalized, automated, and far more dangerous. Ignoring this trend means voluntarily opening the door to the adversary.