Anthropic Cursor Extension Backdoored for 12 Hours

A malicious version of Anthropic's Visual Studio Code extension on Cursor was reportedly backdoored for nearly 12 hours, raising supply-chain concerns for AI coding tools. It remains unclear how attackers gained access to Anthropic's verified Cursor account, according to the report. The episode matters because AI coding assistants often run with broad permissions across repositories and developer environments. A compromised extension could potentially expose source code, credentials or build systems, depending on what access it has. That makes the software supply chain around AI developer tools a high-value target. Developers may install extensions quickly, trust verified publisher badges and grant permissions needed for productivity. Attackers can exploit that trust if an account or release process is compromised. The incident also highlights that AI coding tools face the same distribution and account-security risks as other software, even as they add new capabilities. The reported 12-hour window suggests that detection and removal speed are critical, but so are preventive controls such as multi-factor authentication, release signing, permission scoping and monitoring for unexpected changes. For teams using AI assistants, the practical lesson is to review extension permissions, limit access to sensitive repositories and watch for unusual activity. For vendors, the burden is to secure publishing pipelines and respond quickly when a release is suspected of tampering. The full impact may become clearer as more details emerge, but the core concern is already familiar: trust in an AI tool depends on trust in the code and accounts behind it.