Kimsuky Builds Offline AI Stack to Boost Phishing and Malware Operations

North Korea’s Kimsuky hacking group, operating under the Reconnaissance General Bureau, has been assembling an offline artificial intelligence (AI) stack on its own servers, according to a report from South Korean security firm Genians. The setup includes tools for running language models locally, document retrieval, and speech-to-text transcription, indicating the group is in a research and knowledge acquisition phase aimed at integrating AI into its attack workflow.

Genians discovered the activity after months of tracking infrastructure linked to Kimsuky. The firm found no evidence that the group trained its own AI model, but it did find operational tools such as Ollama, GPT4All, and Msty, which were run or configured rather than merely downloaded. GPT4All contained a configured localdocs_v3.db, the database used by its LocalDocs retrieval-augmented generation (RAG) feature, suggesting the group attempted to connect private documents to an AI system. The report also recovered an operator request to analyze a dataset for wallet details, Gmail credentials, and site-registration history, though it could not confirm the request was submitted to an AI service.

Beyond ready-made applications, the infrastructure hosted developer libraries including LLaMaSharp, Microsoft’s Semantic Kernel, and Microsoft.Agents.AI, which are used to build AI functions into custom C# and .NET software. OpenAI’s Whisper speech-to-text files and a guide on extracting text from audio were also found, along with active traces of Cursor, an AI-powered coding editor. These tools are not exotic, but their deliberate assembly by a nation-state espionage group marks a significant step toward automating parts of its operations.

The activity extends a Kimsuky campaign called Operation GitPower, which abuses GitHub repositories as command channels in an LNK-to-PowerShell infection chain and has distributed encrypted AsyncRAT payloads disguised as image files. Fortinet separately documented the broader GitHub-C2 pattern in April, targeting South Korean users. The newly observed offline AI stack has not been shown running against a victim, and no victim count has been disclosed.

Genians ties the operation to Kimsuky using overlaps with earlier campaigns, infrastructure clues, and North Korean vocabulary recovered from operator logs. The U.S. Treasury sanctioned Kimsuky in 2023, describing it as subordinate to the Reconnaissance General Bureau and focused on intelligence collection. The report also notes a 2025 spear-phishing attack that used ChatGPT-generated images of South Korean military employee ID cards, fitting the pattern of AI adoption.

For defenders, the shift means traditional phishing indicators like stilted translation or spelling mistakes may weaken as AI generates more convincing lures. Genians advises correlating LNK execution, PowerShell, hidden scheduled tasks, GitHub traffic, and later payload activity rather than judging a lure mainly by its polish.