The AI-powered code editor Cursor has confirmed a significant security breach that resulted in the distribution of malicious software through its marketplace. The incident, which occurred on October 29, involved the compromise of credentials belonging to the AI infrastructure provider Fireworks AI. These credentials allowed unauthorized actors to access internal Cursor systems, leading to the exposure of sensitive customer data and the deployment of trojanized versions of the Claude Code extension.
According to the official incident report, the breach originated when attackers obtained an access key stored on Fireworks AI, a third-party service used by Cursor for model training and inference. This key provided access to a read-only storage bucket containing historical AI coding-agent traces. While the majority of these traces were from 2025, they included data from users who had not enabled Privacy Mode. In some instances, these traces contained unredacted API credentials, which the attackers subsequently used to pivot into other Cursor internal systems.
Cursor has confirmed that the breach resulted in the unauthorized access of several internal resources. This included non-production source code repositories, customer support communications, product usage metrics, and contact information. Furthermore, the attackers accessed credentials that had been shared with Cursor’s support team. The company estimates that fewer than 0.05% of its user base had coding-agent traces exposed, and less than 0.01% of customers had their API credentials compromised.
The most critical phase of the attack occurred on October 29, when the intruders utilized a stolen credential to inject malicious versions of the Anthropic Claude Code extension into the cache serving Cursor’s marketplace. The affected versions, identified as 2.1.312 and 2.1.313, were available for download for approximately 11 hours before being removed. Cursor clarified that the legitimate Anthropic extension and the Open VSX registry remained unaffected, as the compromise was limited to the specific cache used by the Cursor editor.
The malicious extensions were modified to execute JavaScript designed to establish persistent SSH backdoors on the host machines of developers. The malware attempted to harvest system information and install an attacker-controlled SSH key to facilitate remote access. However, technical analysis suggests that coding errors within the malicious script likely hindered the backdoor’s functionality on many Windows and macOS systems, particularly those where SSH was not already configured. Cursor estimates that up to 5,000 users may have received the compromised builds during the exposure window.
In response to the discovery, Cursor’s monitoring systems triggered an investigation that led to the revocation of the compromised Fireworks access key and the rotation of all exposed credentials. As a precautionary measure, the company temporarily disabled several internal services, rebuilt its extension marketplace, and redeployed its production infrastructure. Cursor has explicitly stated that its official application releases, production infrastructure, and the code or prompts of users who utilized Privacy Mode were not impacted by the breach.
This incident is part of a broader trend of security threats targeting AI development platforms, a phenomenon recently categorized by security researchers as "Black October." During this period, multiple companies—including Railway, Northflank, Lightning AI, OpenCode, and Warp—reported similar attacks aimed at stealing authentication secrets. These campaigns have frequently resulted in unauthorized AI inference, leading to significant fraudulent charges for affected customers and, in some cases, attempts at AI model distillation.
Fireworks AI itself has disclosed multiple security incidents throughout October, involving both exposed API tokens and unauthorized access to development environments via third-party cloud providers. These breaches have had a cascading effect, impacting other organizations such as Notion, which reported the exposure of AI chat transcripts. Cursor has since moved its inference traffic away from Fireworks AI and is advising all users who installed the affected Claude Code versions to follow its specific remediation guidance.





