Cursor Supply Chain Attack Traced to Fireworks AI Compromise
Share
A sophisticated supply chain attack, originating from multiple compromises of the AI inference platform Fireworks AI, led to the exfiltration of credentials that allowed threat actors to push malicious versions of the Claude Code extension to users of the Cursor VS Code Extension Marketplace.
Developer tool provider Cursor, which integrates large language models into coding workflows, confirmed that its systems were affected after an access key issued to Fireworks AI was compromised. The incident, discovered on 29 October, resulted in at most 5,000 Cursor users potentially installing or updating to tampered versions of the extension.
Fireworks AI Compromise Initiated Attack Chain
The attack on Cursor began with two separate compromises of Fireworks AI in October. Initially, internal Fireworks AI credentials were used to access stored secrets, account, and billing information. A subsequent incident involved the compromise of a third-party cloud provider, which allowed attackers to reach Fireworks AI development machines and steal customer data.
During one of these breaches, a Cursor access key, provided to Fireworks AI, was compromised. Attackers exploited this key to copy data from a read-only storage bucket. This data included unredacted credentials that granted them access to various Software-as-a-Service (SaaS) tools used by Cursor.
Cursor Systems and Data Affected
Leveraging the compromised SaaS tool access, the attackers accessed specific archived non-production Cursor source code repositories, some of which contained business contact information. They also gained access to limited customer support data and email conversations, product usage metrics, contact details, summaries of some coding agent conversations, and certain credentials that users had submitted to the Cursor support team.
Malicious Extension Distribution and User Impact
Although Cursor immediately blocked the attackers’ access upon discovery of the unauthorised activity, the exfiltrated secrets were used to place malicious, unauthorised copies of Anthropic’s Claude Code extension directly onto the cache that serves the Cursor VS Code extension marketplace.
The tampered versions, 2.1.312 and 2.1.313 of the Anthropic.claude-code extension, were available for approximately 11 hours. Cursor noted that the malware was never present in the Open VSX marketplace source itself but was delivered to users who requested the cached versions.
Remediation and Wider Implications
Cursor has urged affected customers who installed or updated to the malicious versions to follow the steps outlined in its official security notice, warning that simply uninstalling the extension may not be sufficient. The company has not observed any attacker activity in its environment since 1 November and considers the incident contained.
The compromise of Fireworks AI had broader implications, as other companies also disclosed data breaches stemming from the incident. Both Notion and Uber have also acknowledged that customer data was compromised due to the Fireworks AI breach, highlighting the significant risks associated with software supply chain security.




Leave a Reply