Cursor Security Breach Update: Key Leak Led to Malicious Extension Injection

The AI-powered code editor Cursor has concluded a two-week investigation into a security breach that resulted in the distribution of a compromised software extension. The incident, which occurred on October 29, was traced back to a leaked access key from Fireworks, a third-party service provider. While the majority of the platform's user base remained unaffected, the company confirmed that a specific subset of developers may have interacted with malicious code during an 11-hour window.

Origins of the Fireworks Access Key Leak

The investigation revealed that attackers initially gained entry by obtaining an access key issued to Fireworks. This credential leak allowed unauthorized actors to copy limited programming agent trace data and subsequently infiltrate various SaaS tools. Through this lateral movement, the attackers managed to access archived source code and internal support data. The breach highlights the growing risks of supply-chain attacks within the AI and development tool ecosystem.

Malicious Claude Code Extension Deployment

Using the stolen credentials, the attackers uploaded a compromised version of the Anthropic Claude Code extension to the Cursor VS Code extension marketplace cache. The specific malicious iterations were identified as:

• Version 2.1.312
• Version 2.1.313

The malicious software was active for approximately 11 hours before the Cursor security team identified the threat and removed the files. According to the company's metrics, up to 5,000 users may have downloaded or updated to the affected versions during this timeframe.

Mitigation and User Safety Measures

In response to the incident, Cursor has implemented several security upgrades to prevent future credential-based attacks. Affected users have been directly notified via email with instructions on how to secure their local environments and rotate sensitive keys.

The vast majority of users were unaffected by this incident, and we have taken immediate steps to harden our infrastructure and marketplace verification processes.

As the integration of LLMs (Large Language Models) and AI agents becomes standard in blockchain development and software engineering, this incident serves as a reminder of the necessity for robust API key management and multi-factor authentication across all development sub-processes.