Malicious Claude Code extensions slipped into Cursor’s marketplace
Attackers published malicious copies of Anthropic’s Claude Code extension (versions 2.1.312 and 2.1.313) on Cursor’s VS Code extension marketplace on October 29, and the tainted packages remained available for roughly 11 hours before removal.
How the breach unfolded
Cursor’s investigation traced the intrusion back to stolen credentials from its third‑party AI infrastructure provider, Fireworks AI. Using those keys, the adversaries accessed a read‑only storage bucket that held historical AI coding‑agent traces. Some of those traces contained unredacted API credentials, giving the attackers a foothold to upload the malicious extensions.
The breach exposed less than 0.05 % of users’ coding‑agent traces and less than 0.01 % of API credentials. Cursor estimates that up to 5,000 developers may have downloaded the compromised extensions during the window of exposure.
Who’s in the cross‑hair
The incident primarily impacts developers who use Cursor’s AI‑powered code editor or the Claude Code VS Code extension from the Cursor marketplace. Because the malicious packages were signed and distributed through the official channel, any downstream users of the extension were potentially exposed.
Cursor also confirmed that the attackers accessed archived, non‑production source‑code repositories and limited customer‑support information, although no evidence suggests production code was compromised.
Response and mitigation
Cursor acted quickly: it revoked the compromised Fireworks AI access key, rotated all exposed credentials, and rebuilt the extension marketplace from scratch. The company also notified affected users and advised them to rotate any API keys that might have been leaked.
Organizations should treat this as a supply‑chain incident. Immediate steps include:
- Audit third‑party credential usage and enforce least‑privilege access.
- Rotate all API keys that were stored in development or CI environments.
- Implement monitoring for unexpected extension uploads in internal marketplaces.
- Educate developers to verify extension provenance, especially for AI‑generated tools.
Broader context
This breach is part of what security firm Wiz calls “Black October,” a series of attacks that have hit companies such as Railway, Northflank, Lightning AI, OpenCode, Warp, and Fireworks AI.
