THE CRUNCH
A supply chain attack on Cursor began with the compromise of AI inference platform Fireworks AI, according to Cursor's account of the incident. Attackers stole a Cursor access key issued to Fireworks and used it to pull data from a read-only storage bucket, where unredacted credentials gave them entry to SaaS tools used by Cursor. From those tools, they reached select archived non-production source code repositories, some customer support records and email threads, product usage metrics, summaries of certain coding agent conversations, and credentials some users had shared with Cursor's support team.
The attack, discovered on October 29, saw two tampered builds of the anthropic.claude-code extension, versions 2.1.312 and 2.1.313, served through the Cursor VS Code Extension Marketplace for roughly 11 hours. The malicious code never appeared in the Open VSX marketplace source; instead it was placed on the cache used to serve requested extensions. Cursor says at most 5,000 users were affected.
The chain started with Fireworks being hacked twice in October: first internal credentials were used to reach Fireworks' stored secrets plus account and billing details, then a third-party cloud provider was breached, giving access to Fireworks development machines and letting customer data be taken. Anthropic was never hacked, and all malicious activity occurred in Cursor's environment.
Cursor says it blocked the attacker's access immediately after discovering the unauthorised activity and considers the incident contained, with no attacker activity seen in its environment since November 1. It warns that anyone who installed or updated to the tampered extension versions during the exposure window should follow the steps in its security notice, as uninstalling the extension alone may not be enough. Notion and Uber have also disclosed data breaches stemming from the Fireworks compromise.


