Build radar
Sources 33Updates always the freshest
0 Newsletter
Animated illustration for: GitHub wants to stop malicious packages before they reach your code
AI News · Oct 21, 2026 · 2 min

GitHub wants to stop malicious packages before they reach your code

GitHub is testing a gate that stops suspicious packages of other people's code before they reach your project. I explain how it works and who can use it.

In short
  • GitHub launched a public preview of a "gate" for downloaded packages that checks them for malicious activity.
  • It supports only npm packages for now and is open to GitHub Enterprise Cloud customers.
  • It is free during the preview.

GitHub announced package firewall in public preview on 20 October 2026. In short: it is a “gate” that downloaded packages of other people’s code pass through before they reach a programmer’s computer, a coding agent or a server that builds an app. According to the post the firewall sits between such endpoints and upstream package registries, and inspects metadata requests and downloads for signs of malicious activity.

How it works

Administrators set the policy. They can choose a cooldown period, meaning how long new packages have to wait, checks against the GitHub Advisory Database for malware and vulnerabilities, and supply-chain checks such as detecting an attestation downgrade. Endpoints also have to be configured to send traffic through the firewall instead of directly to the registry.

Who it is for and what it costs

It is available to GitHub Enterprise Cloud customers, including data-resident ones, and supports only npm for now. It is free during the preview; after general availability GitHub says it will be billed by metered usage, similar to Actions. A package blocked by mistake can be released through an exception request.

How to read it

Coding agents and automated build systems install packages without a human looking, so a filter at the registry level makes sense. It is still a preview, for one platform and one kind of package, so it does not replace ordinary checks. It works only when all traffic really goes through the firewall, and a download that bypasses it will not be checked. So check whether your agents and build servers are connected to it. What to check in code written with AI I describe in the article Security of AI-written code: five checks and in the AI code security hub.

Sources

AI code security

Related