Stolen Vendor Key Led to Trojanized Claude Code Extension on Cursor Marketplace
Cursor has explained how a trojanized version of Anthropic's official Claude Code extension reached its extension marketplace. The company says an attacker obtained an access key it had issued to its vendor, Fireworks AI. Fireworks exposed the key, and the attacker used it to copy data from a read-only storage bucket that Cursor used to share data with the vendor.
The bucket held limited coding agent traces from sessions that were not in Privacy Mode, most dating back to 2025. Some traces contained credentials that had not been removed, exposing the API credentials of less than 0.01% of customers. The attacker used these as a starting point to reach select SaaS tools used by Cursor.
On October 29, 2026, the attacker used a stolen credential from Cursor's internal data to place trojanized Claude Code extension builds in the cache that serves the marketplace. The malicious update tried to install a persistent SSH backdoor. Cursor says at most 5,000 users who requested specific cached versions received the malicious build, which was available for about 11 hours. Affected users installed version 2.1.312 or 2.1.313.
Cursor deactivated the vendor's key, rotated credentials, took the marketplace offline and worked with the hosting provider to remove the attacker's server. It has moved all inference traffic off Fireworks and cut the vendor's access to shared data sets. It reports no further malicious activity since November 1, 2026, and no evidence that Privacy Mode users, customer code repositories, its production environment or other marketplace extensions were affected. The Fireworks AI breach has also hit Uber, which lost driver's license data, and Notion, which confirmed data theft.
Why it matters
This shows how a single supplier's exposed key can cascade into a software supply chain compromise, because a read-only share of data still held live credentials. Businesses using AI coding tools are exposed twice: secrets in session traces can leak through a vendor, and trusted extension updates can become the delivery route for a backdoor. It breaks the assumption that an official extension from a reputable publisher is safe to auto-update.
What to do
- Check whether Cursor installed Claude Code extension version 2.1.312 or 2.1.313 on any developer machine.
- Look for unexpected SSH authorised keys or persistent SSH access on systems that ran the affected versions.
- Rotate any API credentials that were used in Cursor sessions outside Privacy Mode.
- Enable Privacy Mode in Cursor for staff working with sensitive code or credentials.