PromptMink Tricked AI Agents into Planting Malware

#### Executive summary ReversingLabs uncovered the PromptMink campaign, attributed to North Korean-linked Famous Chollima, where attackers use LLM-assisted malicious npm packages (e.g., @validate-sdk/v2) to trick AI coding agents into adding dependencies that steal crypto and secrets. The campaign uses a layered lure/payload model and rapidly evolving payloads (obfuscated JS → SSH key drops → large single-executable malware → Rust-based exfiltration), demonstrating sophisticated supply-chain abuse and LLM optimization; researchers recommend detailed static analysis of dependencies and integrating threat intelligence into developer/AI toolchains.