A criminal group compromised roughly 3,800 internal repositories through a trojanised code editor extension, harvesting code platform credentials, cloud keys, package registry credentials and configuration files from assisted coding tools. Confirmed victims included artificial intelligence companies.
The chosen path describes well where development security is loosest. Editor extensions are installed by individual decision, without passing through review, and run with the same access as the developer, including active production credentials.
Including AI tooling configuration files in the target list is the new detail and deserves attention. Those files typically hold interface keys, internal service addresses and instructions about which systems the assistant may reach.
The number of repositories affected follows naturally from that kind of access. A credential harvested on a developer's machine usually works across several projects, and almost never has a short expiry.
The defence that works is familiar and little practised: short-lived credentials, review of what each installed extension can reach, and a prohibition on storing keys in configuration files inside the repository.