Cursor links extension compromise to exposed Fireworks key

Cursor said in a November 13 incident update that an access key exposed by Fireworks AI let an attacker copy historical coding-agent traces, steal credentials and distribute malicious Claude Code extension builds through Cursor's marketplace cache. Less than 0.01% of customers had API credentials exposed. The traces came from sessions without Privacy Mode, mostly from 2025. Credentials in that material opened access to select internal software tools. Cursor moved all inference traffic off Fireworks, revoked its dataset access and rotated exposed credentials. At most 5,000 users potentially received malicious builds during approximately 11 hours of availability on October 29. Cursor said production infrastructure, customer code repositories and Privacy Mode prompts and code were unaffected.