Security Breach at Cursor Platform Leads to Distribution of Malicious Extensions
The Cursor platform, an AI-powered code editor, confirmed it suffered a security breach that allowed attackers to access internal systems and distribute malicious versions of the Claude Code extension via its extension marketplace. The attack began on October 29, when attackers exploited a leaked access key belonging to Fireworks AI, a third-party provider that Cursor relies on for training AI models.
Details of the Breach and Affected Data
According to the security report issued by the company, the leaked access keys enabled the attackers to copy data from a storage container that contained historical logs of AI coding agents. These logs included sensitive data, including unprotected API Credentials, which allowed the hackers to access Cursor's internal systems and steal various data, including:
- Archived software repositories not intended for production.
- Limited customer support information and email correspondence.
- Product usage metrics and customer contact details.
- Summaries of AI coding agent conversations.
Cursor clarified that less than 0.05% of users were affected by the agent log leak, while the credentials of less than 0.01% of customers were compromised.
Distribution of Malicious Extensions
The attackers used their stolen privileges to publish malicious versions of the Claude Code extension (versions 2.1.312 and 2.1.313) directly into the cache serving the Cursor extension marketplace. These versions remained available for approximately 11 hours before the company discovered and removed them. These versions contained malicious code intended to install SSH Backdoors on developers' machines to ensure remote access, although software bugs in the malicious code may have limited its effectiveness on many operating systems.
What This Means for Users
Cursor estimates that up to 5,000 developers may have installed or updated the extension during the attack window. In response, the company took precautionary measures that included:
- Revoking the access key for Fireworks AI and rotating all affected credentials.
- Rebuilding the extension marketplace and redeploying production infrastructure.
- Moving inference traffic away from Fireworks AI.
This incident is part of a broader series of attacks dubbed "Black October" by security researchers, which targeted AI development platforms to steal credentials and exploit computing resources, underscoring the need to tighten security measures for software supply chains and credentials in cloud development environments.