Back to blog

What breaks when AI floods your dependency list?

Thắng Đoàn
Thắng Đoàn

You install a package the way you always have, and now you are rotating every cloud credential you own. Per Infosecurity Magazine, AI adoption has made this moment likelier by flooding open source with new code faster than anyone can vet it.

AI floods the dependency pool

AI coding tools have removed most barriers to writing and publishing code. Per Infosecurity Magazine, citing a 2026 Stanford report, open-source AI projects on GitHub and Hugging Face hit 5.6 million, triple the 2023 volume. Those same AI tools pull open-source packages indiscriminately, so AI writes the code and AI consumes the code. Infosecurity Magazine also notes about 85% of enterprise software relies on open source, so almost nobody can avoid this surface.

  • 5.6 million open-source AI projects, triple 2023 (Stanford via Infosecurity)
  • About 85% of enterprise software relies on open source, per Infosecurity Magazine

One compromised project, many victims

LiteLLM, a tool many AI stacks depend on, was hit twice in one year. Help Net Security reports its October release pipeline was compromised and pushed v1.107.0 to PyPI and Docker with credential-stealing malware. The malicious package stayed on PyPI from 11:39 to 13:23 UTC. An earlier March breach, analyzed by Hudson Rock, linked a 153GB stolen archive to 2,488 corporate domains, including AWS, Samsung and Cisco.

  • The malicious release harvested cloud logins, developer secrets and AI API keys
  • Impacted releases were pulled, per LiteLLM's blog post

The cheap trick: look-alike packages

ReversingLabs found a fake Material Tailwind NPM package built to pass for the real UI library, per TechTarget. The real library has millions of installations, so a convincing name is enough bait. Attackers do not need a zero-day here. They need a typo.

  • Look-alike package targeted a library with millions of active installations
  • A bigger package pool means more names to impersonate

What this does not solve

Locking dependencies or scanning packages does not fix a compromised release pipeline. LiteLLM's malicious version arrived through the normal distribution channel, so trust in the publisher was the failure point. No source here claims AI writes better malware. The added risk comes from volume and target density, not sophistication. Rotating credentials, as Help Net Security urged, limits blast radius but does not prevent infection.

If your AI coding tools choose packages for you, who reviews that choice?
Which third-party release pipelines do you trust today, and what happens if one is compromised?

List every dependency in your current project and mark which ones auto-update.

Sources

Infosecurity Magazine: AI is Rewriting the Software Supply Chain

Help Net Security: LiteLLM hit by second supply chain attack

TechTarget: Malicious NPM package discovered

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.