Malicious Claude-Generated Commit Targets Crypto Trading Tool

ReversingLabs discovered a supply-chain campaign dubbed PromptMink where a malicious npm package (@validate-sdk/v2) was introduced into an autonomous crypto trading agent, enabling theft of cryptocurrency wallet credentials, system data, project source code, and implantation of SSH keys for persistence. The attackers use a two-layer approach with clean bait packages that import the real payload and employ LLM Optimization (LLMO) tactics—crafting convincing documentation to trick AI coding assistants into recommending malicious dependencies; the activity is attributed to Famous Chollima and defenders are advised to run deep static analysis, enrich AI context with threat intelligence, integrate automated security checks, and manually review AI-generated dependency changes.