DPRK Threat Actors Use Claude Opus to Plant Malicious npm Packages

Cybersecurity researchers uncovered a new wave of DPRK-attributed supply chain attacks using AI-assisted malware insertion. A malicious npm package, “@validate-sdk/v2,” was identified as a dependency because Anthropic’s Claude Opus LLM flagged it as malicious during a code review. The package posed as a utility SDK for hashing, validation, and encoding but contained credential-stealing functionality and a remote access trojan.

The campaign involves DPRK actors operating through fake companies, consistent with Lazarus Group TTPs. Notably, an LLM was used both to generate the malicious package and — inadvertently — to detect it during a developer’s routine AI-assisted code review. Developers should apply automated dependency scanning to all third-party packages before ingestion, and treat AI code review as a signal but not a substitute for dedicated supply chain security tooling.