Cursor confirms security breach led to malicious Claude Code extension
AI code editor Cursor confirmed a security breach that began with stolen credentials from its AI infrastructure provider, Fireworks AI. Attackers accessed a storage bucket containing historical AI coding-agent traces, some with unredacted API credentials, leading to deeper system compromises and the distribution of malicious Claude Code extensions through Cursor’s marketplace.
The breach at Cursor, an AI-powered code editor, originated from an access key stolen from Fireworks AI, a third-party provider for AI model training. This key allowed attackers to copy data from a read-only storage bucket Cursor used to share information with Fireworks. Within this data were historical AI coding-agent traces, primarily from 2025, collected from users who had not enabled Privacy Mode.
A critical aspect of the compromise was that some of these coding-agent traces contained unredacted API credentials. These exposed credentials were then leveraged by attackers to access additional internal Cursor systems and steal data from several of its software-as-a-service (SaaS) tools. This progression illustrates how data intended for AI model training, when not properly secured, can become a vector for broader system infiltration.
The final stage saw attackers use credentials obtained from these further compromised internal systems to distribute malicious copies of Anthropic's Claude Code extension on Cursor's VS Code extension marketplace on October 29. While these malicious extensions, available for about 11 hours, attempted to install persistent SSH backdoors, security researchers noted coding errors that likely prevented them from fully functioning on many Windows and macOS systems where SSH was not already enabled. Cursor estimates up to 5,000 users may have received the malicious builds.
Share this article
Related reading
6 stories
I built an open-source tool that lets AI coding agents collaborate across developers
This piece explores the collaborative potential of AI coding agents, a key element in the Cursor breach.

Self-hosted Claude Managed Agents: what you run, what still leaves your network
We previously examined the security implications of running managed AI agents, a core concern in this incident.

Vietnamese Travellers Can Now Scan NETS QR Codes to Pay in Singapore

Weixin Pay Sees 84% Rise in Overseas E-Wallet Transactions

BPC Pairs SmartVista with Huawei for Payment Modernisation

