Cursor confirms breach as malicious Claude Code extension slips onto its marketplace

Cursor xác nhận vụ xâm nhập khiến tiện ích Claude Code độc hại phát tán

Cursor has confirmed it was breached through a Fireworks AI access key, allowing a malicious Claude Code extension to reach its marketplace and putting up to 5,000 developers at risk.

⚡ MIGOVI TL;DR

  • Hackers obtained an access key stored on Fireworks AI, then used it to copy Cursor's data and steal additional credentials.
  • On October 29, two malicious Claude Code versions (2.1.312 and 2.1.313) appeared on Cursor's marketplace for about 11 hours.
  • Up to 5,000 users may have received the malicious build that installs an SSH backdoor, though a coding error may stop the backdoor from running on many Windows and macOS machines.
  • The incident is part of an attack campaign dubbed "Black October" by security firm Wiz, targeting AI development infrastructure.

It started with an access key on Fireworks AI

Cursor has confirmed a breach, according to an incident report published Friday. The attacker obtained an access key stored on Fireworks AI, the provider of AI model training and inference infrastructure that Cursor uses. Cursor is an AI-powered code editor built on Microsoft's open-source Visual Studio Code (VS Code).

With this key, the hacker copied data from a read-only storage bucket that Cursor uses to share information with Fireworks. The bucket held traces of AI coding agent activity, mostly from 2025, collected from users who had not enabled Privacy Mode. Some traces also contained unredacted API credentials, which the attacker used to reach additional Cursor systems.

Diagram of the attack on Cursor that led to the malicious Claude Code extension

According to Cursor, fewer than 0.05% of users had coding agent traces exposed, and fewer than 0.01% of customers had API credentials exposed. From the company's SaaS tools, the hacker obtained archived non-production code repositories (some containing business contact information), some customer support data and email threads, product usage metrics, customer contact details, summaries of some coding agent conversations, and credentials that customers had previously sent to the support team.

A fake Claude Code extension on the marketplace

Cursor's monitoring systems detected the unauthorized activity. The company revoked the Fireworks key, rotated the exposed credentials, and temporarily shut down some internal services and products as a precaution. By then, however, the attacker had already obtained a credential that allowed publishing extensions to Cursor's marketplace.

On October 29, the hacker used the stolen credential to place unauthorized copies of the Claude Code extension (from Anthropic) directly into the cache serving Cursor's VS Code extension marketplace. Anthropic's original extension and Open VSX, the marketplace Cursor relies on, were not compromised.

The two malicious versions, 2.1.312 and 2.1.313, stayed up for about 11 hours before Cursor removed them. According to BleepingComputer, they were modified to run malicious JavaScript that installs a persistent SSH backdoor. The malware collects system information, installs an attacker-controlled SSH key, tries to enable SSH, and then sets up a mechanism for persistent remote access. Security researchers say a coding error may prevent the backdoor from working on many Windows and macOS machines that don't already have SSH enabled.

Attack chain diagram showing how the malicious Claude Code extension reached Cursor users

Cursor estimates that up to 5,000 users downloaded the affected versions during that window. Production infrastructure, official application releases, customer code repositories, and the code and prompts of users with Privacy Mode enabled were not affected. The company says it has moved all inference traffic off Fireworks, rotated credentials, rebuilt the extension marketplace, and redeployed production infrastructure. Users who installed or updated the Claude Code extension during the attack should follow Cursor's remediation guidance.

Part of the "Black October" campaign

Security firm Wiz calls the attack part of "Black October," a wave targeting AI developer platforms. The main goal is development infrastructure, with attackers seeking credentials, especially those that grant access to AI models. Stolen credentials are used to run unauthorized inference, leaving victims with large bills. Some incidents also involved attempts at AI model distillation.

Several companies have been affected, including Railway, Northflank, Lightning AI, OpenCode, and Warp. At Railway, the attacker obtained customer authentication keys and generated roughly 740,000 USD in fraudulent AI charges. OpenCode also reported exposed API keys and unauthorized AI usage.

Fireworks AI itself disclosed multiple breaches in October. One involved exposed API keys and tokens, while another saw an attacker access a development machine through a hijacked third-party cloud provider. The incidents also hit Fireworks customers such as Notion, which recently said that AI chat logs from a small number of workspaces were exposed.

MIGOVI VIEWPOINT

The incident shows that supply chain risk no longer lies only in open-source libraries but also in the operational data of AI tools. Coding agent traces can easily contain API keys that users never notice. If you use Cursor, enable Privacy Mode, check which Claude Code version you have installed, and rotate any keys you have pasted into the editor or sent to the support team. For development teams, managing secrets on a short-lived, least-privilege basis will limit the damage when a third-party provider is breached.

Source: BleepingComputer

Reviewed by: Hero Association

Share your thoughts ^^

Your email address will not be published. Required fields are marked *

Insights

Contents

  1. It started with an access key on Fireworks AI
  2. A fake Claude Code extension on the marketplace
  3. Part of the "Black October" campaign