Cursor confirmed that attackers used a key taken from Fireworks AI to copy agent traces from a shared storage bucket. Some traces held unredacted API credentials, which were used to publish two malicious Claude Code extension builds on 29 October. They were live for about 11 hours and up to 5,000 users may have received them. The malware tried to install SSH backdoors but appears to have failed on many systems. Teams using Cursor should check whether the affected versions were installed and follow Cursor's notice, and should rotate any credentials that could appear in agent traces.
