Cursor Traces Breach to Access Key Exposed by Fireworks AI

What happened

Cursor announced that an attacker used an access key exposed by Fireworks AI to obtain historical coding-agent data and credentials. This allowed the attacker to distribute unauthorized, malicious Claude Code extension builds to a small number of users through Cursor’s VS Code extension marketplace. The incident was detected and contained, with affected users notified directly. Cursor rebuilt its extension marketplace and rotated all exposed credentials to prevent further impact.

Key details

• The breach affected fewer than 0.01% of Cursor customers.
• Malicious extensions were available for approximately 11 hours on October 29, 2026.
• Exposed data included coding-agent traces, some unredacted credentials, and limited business contact information.
• Current customer data, live traffic, and official app releases were not impacted.
• Cursor moved all inference traffic off Fireworks AI and revoked the vendor’s access to shared datasets.

Why it matters

Cursor customers who downloaded specific cached versions of the Claude Code extension during the breach window may have received a malicious build, requiring them to review and possibly rotate credentials. Organizations relying on Cursor’s VS Code extensions faced temporary disruption as the marketplace was validated and rebuilt. Cursor and Fireworks AI have both changed their security practices, affecting how vendors and customers manage access keys, credentials, and data sharing going forward.