According to Cursor, the breach started with an access key stored at Fireworks AI, the AI infrastructure provider it worked with. The attackers used it to copy data from shared storage holding historical coding-agent traces, mostly from 2025, from users who had not enabled Privacy Mode. Some traces contained unredacted credentials, which the attackers used to reach more Cursor systems: archived non-production code repositories, support tickets and credentials customers had previously shared with the support team. Cursor says fewer than 0.05% of users had traces exposed and fewer than 0.01% of customers had API credentials compromised. The Fireworks breach itself was covered here earlier.
One of the stolen credentials let the attackers publish extensions. On 29 October they placed versions 2.1.312 and 2.1.313 of Anthropic's Claude Code extension in the cache that serves Cursor's extension marketplace, modified to try to install a persistent SSH backdoor. The versions were available for about 11 hours. Anthropic's legitimate extension and Open VSX were not compromised, and researchers found coding errors that likely stopped the malware from working on Windows and macOS machines where SSH was not already enabled.
Cursor says its production infrastructure, official app releases, customer code repositories and data from Privacy Mode users were not affected. The company moved all inference traffic away from Fireworks, rotated exposed credentials and rebuilt its extension marketplace. The incident is part of a wave of attacks on AI development platforms that security firm Wiz has dubbed "Black October".
What to do: anyone who installed or updated the Claude Code extension in Cursor on 29 October should follow Cursor's remediation guidance, check for an unfamiliar SSH key on the machine, and rotate API keys and credentials stored on it.