Security

Claude Code Extension Attack: SSH Backdoor Findings

3 min read

Summary

Microsoft Threat Intelligence analyzed a malicious Claude Code-branded extension briefly distributed through the Cursor marketplace that attempted to install an SSH backdoor, create reverse tunnels, and establish persistence. The incident appears limited to Cursor marketplace, but organizations should identify affected systems, preserve evidence, remove artifacts, and review endpoint and network telemetry for signs of compromise.

Need help with Security?Talk to an Expert

Introduction

Microsoft has published new research into a malicious Claude Code-branded extension briefly hosted in the Cursor marketplace on October 29, 2026. For security teams and administrators, this matters because the tampered package attempted to create persistent SSH-based remote access while continuing to appear functional to the user.

Importantly, Microsoft says the incident was not tied to Visual Studio Marketplace or the npm @antropic-ai/claude-code package, and current evidence indicates the exposure was limited to the Cursor marketplace.

What happened

According to Microsoft Security Research, the malicious versions were:

  • anthropic.claude-code versions 2.1.312 and 2.1.313
  • Available in the Cursor marketplace for a limited window on October 29, 2026
  • Designed to run injected JavaScript before the extension’s expected code

What the malicious extension did

Microsoft’s analysis found the extension attempted to:

  • Collect host and user reconnaissance data
  • Append an attacker-controlled SSH public key to authorized_keys
  • Enable the local SSH service
  • Create reverse SSH forwarding to expose the local machine through an outbound tunnel
  • Establish platform-specific persistence on Windows, macOS, and Linux
  • Report host details and selected environment-variable names back to the attacker

The malware used built-in tools and embedded components rather than downloading another payload, which can make detection harder. It also suppressed command output and silently handled errors to avoid drawing attention.

Why this matters for IT and security teams

Even if the reverse tunnel did not succeed, the extension may still have:

  • Written SSH authorization keys
  • Created persistence artifacts
  • Collected system metadata
  • Left behind files or scheduled startup mechanisms

That means administrators should not assume failed attacker access simply because no active SSH session is visible. Systems that installed or auto-updated to the affected versions should be treated as potentially compromised until investigated.

Microsoft recommends organizations take these steps:

  • Identify endpoints that installed the affected Cursor marketplace extension versions
  • Preserve forensic evidence before cleanup
  • Remove confirmed malicious keys, scripts, and persistence artifacts
  • Review endpoint and network telemetry for SSH tunneling or suspicious process activity
  • Rotate credentials or revoke sessions where investigation suggests exposure
  • Use Microsoft Defender XDR detections, threat analytics, and advanced hunting guidance once available and validated

Bottom line

This incident is a strong reminder that extension marketplaces can become a supply-chain risk. If your organization uses Cursor, security teams should immediately verify whether any devices received the malicious Claude Code extension and begin containment and remediation where needed.

Need help with Security?

Our experts can help you implement and optimize your Microsoft solutions.

Talk to an Expert

Stay updated on Microsoft technologies

SecuritySSH backdoorCursor marketplacesupply chain attackMicrosoft Defender XDR

Related Posts

Security

AD CS ESC1 and ESC8 Attacks: Microsoft Findings

Microsoft Threat Intelligence detailed two intrusions where attackers used Microsoft Teams help desk impersonation and Quick Assist sessions to gain remote access, then abused Active Directory Certificate Services through ESC1 and ESC8 paths. The research highlights how misconfigured AD CS environments can turn a socially engineered session into durable privileged access, making certificate hardening and containment coverage critical for defenders.

Security

North Korean Threat Actors Expand AI Attack Tradecraft

Microsoft Threat Intelligence says North Korean threat actors are rapidly expanding AI use across reconnaissance, social engineering, tooling development, and operational scaling. The report matters for security teams because AI is accelerating familiar attack chains rather than replacing them, increasing the speed and volume of targeted campaigns against financial, technology, government, and education organizations.

Security

Microsoft Defender Named a Leader in IDC 2026

Microsoft has been named a Leader in the 2026 IDC MarketScape for Worldwide Modern Endpoint Security for Enterprises. The recognition highlights Microsoft Defender’s autonomous attack disruption, generally available agentic AI in Security Copilot, and broad cross-platform coverage, signaling continued investment in faster, AI-driven endpoint protection for enterprise security teams.

Security

CISO Role in Agentic AI: Security Priorities for 2026

Microsoft says CISOs now face a dual challenge in the agentic AI era: governing enterprise AI agents safely while also defending against faster AI-enabled attacks. The guidance emphasizes deep observability, governed identities, data protection, and integrated security operations so organizations can innovate without creating unmanaged risk.

Security

LiteLLM 1.107.0 Breach: Python Hook Credential Theft

Microsoft has detailed a supply chain compromise in LiteLLM 1.107.0 that used Python startup hooks to launch credential theft automatically. The issue is especially serious because simply starting Python in an affected environment could trigger the payload, putting developer workstations, CI/CD agents, containers, and LiteLLM proxy deployments at risk.

Security

ARTEX AI Attacks Drive ConnectWise Supply Chain Breach

Microsoft says China-aligned actor Storm-0940 used the ARTEX agentic AI framework to automate large-scale attacks and accelerate the ConnectWise supply chain compromise. The campaign reached thousands of Microsoft 365 tenants, abused OAuth applications, and in some cases achieved privileged cloud access in under three hours.