A supply chain attack against Cursor users originated from the compromise of Fireworks AI, an AI inference platform. Credentials stolen during the Fireworks AI breach enabled attackers to steal additional secrets and push malicious copies of the anthropic.claude-code extension through the Cursor VS Code Extension Marketplace.
The incident was discovered on October 29. Threat actors published two modified versions of the extension (2.1.312 and 2.1.313) that remained available for roughly 11 hours. Anthropic was never hacked; all malicious activity occurred within Cursor’s environment after a Fireworks AI access key was compromised.
Fireworks AI suffered two breaches in October. First, internal credentials were used to access stored secrets and account information. Later, a third-party cloud provider was compromised, allowing attackers to reach development machines and exfiltrate customer data. The stolen Cursor access key was used to copy data from a read-only storage bucket, exposing unredacted credentials for Cursor SaaS tools.
From these tools the attackers accessed archived non-production source code repositories, limited customer support data, email conversations, product usage metrics, summaries of coding agent conversations, and credentials sent to support. Cursor immediately blocked the unauthorized access, but the attackers used exfiltrated secrets to place malicious extension copies in the marketplace cache. The malware never appeared in the Open VSX marketplace source.
At most 5,000 users were affected. Cursor advises customers who installed versions 2.1.312 or 2.1.313 to follow the steps in its security notice, as simply uninstalling the extension may not be sufficient. The company considers the incident contained, with no attacker activity observed since November 1. Notion and Uber also disclosed breaches stemming from the same Fireworks AI compromise.


