Supply Chain Alert: Malicious Claude Code Extension Hit Developer Machines via Cursor Marketplace
A critical supply chain attack saw compromised versions of Anthropic's Claude Code extension distributed via Cursor's marketplace, affecting up to 5,000 developers with persistent SSH backdoors.

In a stark reminder of the ever-present threats within the software supply chain, thousands of developers found themselves targeted by a sophisticated attack involving a seemingly legitimate AI coding assistant. On October 29, 2026, maliciously modified versions of Anthropic's Claude Code extension were distributed through the Cursor VS Code Extension Marketplace, planting persistent SSH backdoors on affected macOS and Windows machines. This incident, which lasted approximately 11 hours, underscores the critical importance of scrutinizing every component in our development ecosystems, even those from trusted sources.
The attack vector was particularly insidious: a tampered version of a widely used developer tool, delivered through an official marketplace. This bypasses many traditional security checks and highlights the evolving nature of cyber threats targeting the developer community. While the immediate threat was mitigated, the implications for software supply chain security are profound, urging developers and organizations to re-evaluate their security postures and verification processes for third-party tools and extensions.
1. The Anatomy of the Attack: How it Unfolded
The incident began on October 29, 2026, when versions 2.1.312 and 2.1.313 of the Anthropic Claude Code VS Code extension were published to Cursor's VS Code Extension Marketplace. These versions, intended for macOS and Windows users, contained an embedded malicious JavaScript loader within their extension.js bundle. The malicious builds were available for approximately 11 hours, from 2:27 a.m. to 1:35 p.m. PT (09:27–20:35 UTC).
Crucially, this was identified as a supply chain attack on Cursor's marketplace, not a direct compromise of Anthropic's systems or their official distribution channels. A malicious actor gained access to Cursor's marketplace to publish the tampered copies, which Anthropic confirmed they did not build or publish. This distinction is vital, as it means developers relying on Anthropic's official channels or other marketplaces like Microsoft's Visual Studio Marketplace or Open VSX were not affected by this specific incident. To prevent further confusion, Anthropic subsequently updated its legitimate Claude Code extension directly to version 2.1.314.
During the 11-hour window, an estimated 5,000 Cursor users may have downloaded or updated to these compromised versions. The malicious code was designed to execute as soon as the extension module loaded, even before the extension's normal activation callback, meaning user interaction with the extension was not required for the payload to fire.
2. The Malicious Payload: What the Backdoor Did
Upon execution, the injected JavaScript loader initiated a multi-stage attack designed to establish persistent remote access and gather sensitive information. The primary objectives of the malicious payload included:
- Host and User Information Collection: The malware collected details such as the hostname, username, operating system, architecture, home directory path, environment variable names, working directories, configured Git email, and the public IP address of the compromised machine. This reconnaissance data was then sent to an attacker-controlled server.
- SSH Key Installation: An attacker-controlled SSH public key, labeled
svc-telemetry, was added to the user'sauthorized_keysfile. This key would theoretically allow the attacker to log into the victim's machine. - Reverse SSH Tunnel Establishment: The malicious code attempted to enable the local SSH service and open a reverse SSH tunnel to an external IP address, specifically
134.209.82[.]225. This tunnel aimed to provide a persistent communication channel back to the attacker's command-and-control server. - Persistence Mechanism: To ensure continued access, the malware established persistence on the infected systems. On macOS, this involved writing a LaunchAgent (
~/Library/LaunchAgents/com.apple.sysmond.plist). On Windows, it attempted to create a scheduled task named 'SysMon' running as SYSTEM, with a fallback to copyingsysmon.cmdinto the user's Startup folder. These persistence mechanisms would retry the SSH tunnel approximately every 60 seconds.
While the intent was clear, Cursor reported that due to errors in the malicious code and the requirement for administrator privileges to enable the local SSH service, the SSH access step would likely fail on most macOS and Windows hosts. However, even if the reverse tunnel failed, the unauthorized SSH key and persistence mechanisms could still be installed on the system, leaving it vulnerable.
3. Remediation and Recovery: Beyond Uninstalling the Extension
A critical takeaway from this incident is that simply uninstalling the compromised Claude Code extension is insufficient for full remediation. The malicious payload establishes artifacts outside the extension's installation directory, meaning these elements persist even after the extension is removed.
Affected developers who installed or updated versions 2.1.312 or 2.1.313 from the Cursor marketplace during the exposure window must perform a thorough cleanup. Key remediation steps, as advised by Cursor and security researchers, include:
- Identify Impacted Machines: Check for the presence of the
svc-telemetrySSH public key in~/.ssh/authorized_keys(macOS/Linux) or%USERPROFILE%\.ssh\authorized_keysandC:\ProgramData\ssh\administrators_authorized_keys(Windows). - Terminate Malicious Processes: Kill any active SSH tunnel processes or retry loops associated with the attack, especially those connecting to
134.209.82[.]225. - Remove Persistence Mechanisms: Manually remove the malicious LaunchAgent on macOS (
~/Library/LaunchAgents/com.apple.sysmond.plist), cron entries on Linux, or the scheduled task/Startup folder script on Windows. Also, delete associated scripts and identity files (e.g.,~/.config/.smt,~/.config/.smtk). - Clean Up SSH Keys: Remove the unauthorized
svc-telemetrypublic key from yourauthorized_keysfile. It is crucial to confirm the key's fingerprint before removal to avoid disrupting legitimate SSH configurations. - Rotate Credentials: Given the potential for host information collection, it is highly recommended to rotate any sensitive credentials, API tokens, or keys that were present or accessible on the compromised machine.
- Assess and Rebuild: In severe cases, especially where remote access cannot be definitively ruled out or the extent of compromise is unclear, rebuilding the affected host from a clean image should be considered.
Cursor took immediate action by removing the malicious versions, temporarily shutting down its marketplace, rotating credentials, rebuilding its infrastructure, and working with cloud providers to take the attacker's servers offline.
4. Broader Implications for Software Supply Chain Security
This incident serves as a critical case study in the ongoing challenges of software supply chain security. As modern development increasingly relies on a complex web of third-party libraries, tools, and extensions, the attack surface expands dramatically. The fact that a trusted AI coding assistant, delivered via an official marketplace, could be weaponized highlights several key concerns:
- Trust in Marketplaces: Developers often implicitly trust extensions and tools distributed through official marketplaces. This attack demonstrates that even these channels can be compromised, necessitating more robust vetting and continuous monitoring by marketplace operators.
- The 'Trusted Publisher' Fallacy: The malicious extension was distributed under Anthropic's legitimate identifier, making it indistinguishable from a genuine update to the unsuspecting developer. This type of impersonation challenges traditional trust models.
- Beyond Code Vulnerabilities: This was not an exploit of a vulnerability within the Claude Code extension itself, but rather an abuse of the distribution mechanism. Supply chain security must encompass the entire software development lifecycle, from source code to delivery.
- Endpoint Security for Developers: Developer workstations are prime targets due to the sensitive code, credentials, and access they often hold. Robust endpoint detection and response (EDR) solutions, alongside secure coding practices and continuous monitoring, are essential.
Organizations are urged to adopt comprehensive software supply chain security practices, including establishing vendor risk management, leveraging Software Bill of Materials (SBOMs) to track components, and continuously monitoring for new vulnerabilities. The CISA (Cybersecurity and Infrastructure Security Agency) provides recommended practices for developers to enhance software integrity and security.
Comparison Overview
| Feature/Item | Legitimate Version (e.g., 2.1.314) | Malicious Version (2.1.312, 2.1.313) |
|---|---|---|
| Distribution Channel | Anthropic's official channels, Microsoft VS Code Marketplace, Open VSX | Cursor VS Code Extension Marketplace ONLY |
| Publisher | Anthropic | Malicious actor using Anthropic's identifier |
| Payload | Standard AI coding assistant functionality | Embedded JavaScript loader for backdoor |
| Impact on System | Aids in coding tasks | Collects data, installs SSH key, creates persistence, attempts reverse tunnel |
| Remediation | Standard uninstall | Manual removal of SSH keys, persistence, and potential system rebuild required |
| Affected Users | All users | Up to 5,000 Cursor users on macOS/Windows during specific window |
Frequently Asked Questions (FAQ)
Q: What is a software supply chain attack?
A software supply chain attack targets vulnerabilities in the software development process or its components. Instead of directly attacking a target organization, attackers compromise a third-party component, tool, or distribution mechanism that the target uses, thereby gaining unauthorized access or distributing malware through trusted channels.
Q: How do I know if I was affected by the Claude Code extension incident?
You might be affected if you are a Cursor user on macOS or Windows and installed or updated the Anthropic Claude Code extension (versions 2.1.312 or 2.1.313) from the Cursor marketplace between October 29, 2026, 2:27 a.m. and 1:35 p.m. PT. The most definitive check is to look for an attacker-controlled SSH public key labeled 'svc-telemetry' in your system's SSH authorized_keys file.
Q: Is uninstalling the malicious extension enough to fix the problem?
No, simply uninstalling the extension is not enough. The malicious code establishes persistence mechanisms and plants artifacts outside the extension's installation directory. You must manually remove the unauthorized SSH keys, persistence entries (like LaunchAgents or scheduled tasks), and potentially rebuild your system to ensure full remediation.
Q: What steps can developers take to protect against similar supply chain attacks?
Developers should implement secure coding practices, establish robust vendor risk management for third-party tools, adopt open-source software management practices, leverage Software Bill of Materials (SBOMs) to track components, and continuously monitor for vulnerabilities. Additionally, exercising caution with new extensions, verifying sources, and utilizing strong endpoint security are crucial.
Try Our Developer Utilities
Simplify your engineering workflows with our free browser-native tools: