The Hacker News
Stolen Fireworks AI key led to trojanized Cursor extension
Cursor says a stolen Fireworks AI access key led to the compromise of its Claude Code extension on Cursor Marketplace, following a breach that exposed Fireworks credentials. Cursor reported that an attacker obtained a key it had issued to Fireworks, then used those credentials to copy data from a read-only storage bucket it used to share data with Fireworks.
The bucket reportedly contained limited traces from sessions not in Privacy Mode, including some credentials that were not scrubbed, exposing API credentials for less than 0.01% of Cursor’s customers. The attacker then used the exposed data to access select SaaS tools used by Cursor and, on 29 October 2026, to place trojanized versions of Anthropic’s Claude Code extension in the cache used to serve Cursor’s extension marketplace.
Up to around 5,000 users who requested cached versions during the exposure window could have received the malicious build, which Cursor removed after about 11 hours and took the marketplace offline.
Cursor says it moved inference traffic off Fireworks, deactivated the compromised key, rotated credentials, and notified affected users who had installed versions 2.1.312 or 2.1.313 from the marketplace during the window. It also revoked exposed credentials and cut off access to data sets it had shared with Fireworks.
Cursor asserts that there has been no evidence of further malicious activity since 1 November 2026 and that the incident did not affect Privacy Mode users, Cursor’s production environment, or other Cursor Marketplace downloads. Separately, Uber confirmed that threat actors stole thousands of drivers’ licences after compromising a service provider’s internal credentials, a breach linked to Fireworks AI, though Uber did not disclose numbers.