AI Pulse · AI Trends Pulse

A supply-chain worm hit npm last week, and this one was built to hunt AI coding assistants. Someone compromised a maintainer account and pushed malicious code into keyv and three related packages that collectively see 127 million downloads a week. The malware grabs cloud credentials, then copies itself forward into every project that installs the poisoned package.

What makes this different is the targeting. The payload drops autostart hooks into .claude and .vscode folders, so the malicious code fires the moment you open a repo in Claude Desktop or VS Code. If your team uses Cursor, Windsurf, or any editor that leans on those assistants, the worm is sitting in the workflow, not just the build pipeline. The original report confirms 868 packages and 1,381 versions were compromised as of August 4.

What you do now

Audit your package-lock.json for keyv, cacheable, flat-cache, and file-entry-cache. If you see install dates between late July and early August, assume compromise and rotate credentials. Check .claude and .vscode folders in every repo your team touches. If you find preinstall or postinstall scripts you did not write, delete them and re-clone clean.

Longer term, this is why dependency pinning and private registries matter. If your build pulls the latest version of anything automatically, you are one compromised maintainer away from handing an attacker your AWS keys. The kind of guardrails we build into the Omni Command Centre include dependency scanning and credential vaults that do not live in plaintext on developer machines, so a worm that reads environment variables hits a wall instead of your production account. You do not need to solve this with process documents. You need infrastructure that assumes compromise and limits the blast radius.