Cursor Confirms Security Breach Led to Malicious Claude Code Extension Distribution
Cursor has confirmed a significant security breach that allowed attackers to distribute a malicious Claude Code extension through its marketplace. The incident originated from credentials stolen from AI infrastructure provider Fireworks AI, compromising internal systems and customer data.
Table of Contents
The breach exposed up to 5,000 developers to malware designed to install persistent SSH backdoors. This event highlights the growing risks associated with the interconnectedness of AI development tools and infrastructure.
.
Attack Vector: Compromised Fireworks AI Credentials
The attack commenced when threat actors obtained an access key from Fireworks AI, a crucial third-party provider for AI model training and inference. This key enabled attackers to access a read-only storage bucket used by Cursor to share data with Fireworks.
Within this bucket were historical AI coding-agent traces, primarily from 2025, collected from users who had not enabled Privacy Mode. Alarmingly, some of these traces contained unredacted API credentials, which the attackers leveraged to gain further access to Cursor’s internal systems.
Data Exfiltration and SaaS Compromise
Using the pilfered credentials, the hackers exfiltrated data from several of Cursor’s Software-as-a-Service (SaaS) tools. This included non-production source code repositories, limited customer support information, product usage metrics, and summaries of AI coding-agent conversations.
Crucially, credentials previously shared by customers with Cursor’s support team were also compromised. Cursor estimates that fewer than 0.05% of users had their coding-agent traces exposed, and an even smaller fraction, less than 0.01%, experienced API credential compromise.
Malicious Extension Deployment on Cursor Marketplace
Despite Cursor’s swift response in revoking the Fireworks access key and rotating exposed credentials, the attackers had already secured a credential enabling them to publish extensions via Cursor’s marketplace. On October 29, malicious, unauthorized copies of Anthropic’s Claude Code extension were published.
These malicious versions, 2.1.312 and 2.1.313, remained available for approximately 11 hours before being removed. The attack specifically targeted the Claude Code extension, modifying it to execute malicious JavaScript aimed at establishing persistent SSH backdoors.
.
Impact on Developers and Remediation
The malware attempted to collect system information, install attacker-controlled SSH keys, and enable persistence mechanisms for remote access. While coding errors may have limited the backdoor’s effectiveness on some systems, Cursor estimates up to 5,000 users may have downloaded the compromised versions.
Cursor assures that its production infrastructure, official application releases, and user code with Privacy Mode enabled were unaffected. The company has since moved inference traffic away from Fireworks, rotated all compromised credentials, and rebuilt its extension marketplace.
Broader Context: “Black October” and AI Security Risks
This incident is part of a broader trend of attacks targeting AI developer platforms, a phenomenon cybersecurity firm Wiz has termed “Black October.” These attacks focus on stealing authentication credentials to gain unauthorized access to AI models and inference capabilities, often resulting in significant fraudulent charges for affected customers.
Companies like Railway and OpenCode have previously reported similar breaches involving stolen secrets and unauthorized AI usage. Fireworks AI itself disclosed multiple incidents in October, underscoring the vulnerability of the AI supply chain. This situation echoes concerns about the security implications of AI code generation, as discussed in our guide on AI Writes Code, Engineers Own the System.
Developers who installed or updated the affected Claude Code extension versions are strongly advised to follow Cursor’s remediation guidance to secure their systems against potential threats.
Source: BleepingComputer Security


