Dev.to Security πŸ” Cybersecurity πŸ‘ 0 πŸ“– 2 min read

Your AI Agent Just Installed a Package. Did Anyone Check It?

TL;DR AI coding agents install dependencies on their own, and most supply chain defenses assume a human picked each one. Xygeni just won two Top InfoSec Innovator Awards 2026 from Cyber Defense Magazine for closing tha

TL;DR

AI coding agents install dependencies on their own, and most supply chain defenses assume a human picked each one. Xygeni just won two Top InfoSec Innovator Awards 2026 from Cyber Defense Magazine for closing that gap. Here’s the problem and how we approach it.

A familiar session, with one detail missing

agent: Adding CSV export to the reports module
agent: Installing dependency...
$ npm install fast-csv-export
added 1 package in 2s
agent: Tests passing. Opening pull request.

Looks fine. But nobody in that loop asked:

  • How old is that package?
  • Who published it?
  • What did its install script just do on this machine?

The agent didn’t ask because it can’t. And you didn’t ask because you weren’t there. That’s the point of the agent.

(The package name above is made up, which is exactly the problem: agents regularly suggest packages that don’t exist, and attackers register those names.)

Why the usual checks miss it

Most supply chain tooling asks some version of β€œis this package known to be bad?” Against a package published this morning, the answer is no, and β€œno” doesn’t block anything. No signature, no CVE, no reputation history.

What we do about it

We built Xygeni to act at the moments where code is actually created:

πŸ€– Inside the agent. Coding agents call Xygeni through its MCP server, scan what they write and get fixes, under the same policies as human code.
πŸ“¦ At publication. Malware Early Warning analyzes new versions on npm, PyPI, Maven and other registries as they appear, and flags malicious packages before a signature exists.
πŸ’» On your laptop. Shield blocks malicious packages before they download, so install scripts never run. You can also require a minimum age for new packages, extensions and plugins.
πŸ—ΊοΈ Across your AI stack. AI Security inventories the models, agents, MCP servers and skills in your repos.

That’s what Cyber Defense Magazine recognized this week, with awards in Most Innovative GenAI Agentic Application Security and Transformational Software Supply Chain Security.

Try it yourself

Malware Early Warning is included in the free Developer plan, with no credit card

Full announcement

How are you handling dependencies your agents add? Curious what’s working for other teams. πŸ‘‡

πŸ“° Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.