VS Code

Cursor Confirms Breach via Fireworks AI Exposed Customer Data and Enabled Malicious Claude Code Extension

Cursor disclosed a security incident that began with stolen credentials from Fireworks AI, allowing attackers to access internal systems and distribute malicious versions of Anthropic's Claude Code extension. The breach exposed historical AI coding-agent traces from users without Privacy Mode enabled, including unredacted API credentials that facilitated further compromise of SaaS tools. Attackers published malicious extension versions 2.1.312 and 2.1.313 on Cursor's marketplace for about 11 hours, affecting up to 5,000 developers with attempts to install persistent SSH backdoors. Cursor revoked the Fireworks access key, rotated credentials, rebuilt its extension marketplace, and moved inference traffic away from the provider. The incident forms part of a broader wave of attacks on AI developer platforms known as Black October, which also impacted Railway, Northflank, and others with credential theft and unauthorized AI usage. Production infrastructure and Privacy Mode data remained unaffected according to the company.