Trojanized Claude Code extension served through Cursor marketplace installs SSH backdoor

Cursor has disclosed a supply-chain compromise in which malicious copies of the Claude Code extension were distributed to users of its AI code editor through the Cursor extension marketplace. The tampered builds, published under the genuine anthropic.claude-code extension ID as versions 2.1.312 and 2.1.313, were available for roughly 11 hours on 29 October 2026 and could have reached up to 5,000 users.

What happened

Cursor’s 13 November update explains the chain. An attacker obtained an access key that Cursor had issued to inference provider Fireworks, which Fireworks had exposed. With it, the attacker copied a read-only storage bucket containing mostly 2025 coding-agent traces, some of which held unredacted credentials. Those credentials opened access to selected SaaS tools and, ultimately, to the cache that serves Cursor’s VS Code extension marketplace, where the attacker placed the malicious copies.

Anthropic did not build or publish the tampered extension. Microsoft confirmed the malicious packages never reached the Visual Studio Marketplace or the @anthropic-ai/claude-code npm package, and Open VSX, which Cursor’s marketplace mirrors, was never affected. A clean 2.1.314 was published on 30 October.

Technical details

Analyses by Microsoft, Sophos, StepSecurity and others describe Windows and macOS payloads that add an attacker-controlled key to authorized_keys, attempt a reverse SSH tunnel to 134.209.82[.]225, establish persistence and report host details. Cursor also says a very small share of customers had agent traces or API credentials exposed via the bucket.

Who is affected

Cursor users who installed or auto-updated the Claude Code extension on 29 October. Users of the official VS Code marketplace or npm package were not affected.

Recommendations

  • Check installed extension versions; if 2.1.312 or 2.1.313 was present, follow Cursor’s security notice and run its check scripts. Uninstalling the extension alone may not remove persistence.
  • Inspect ~/.ssh/authorized_keys for unknown keys and block egress to 134.209.82.225.
  • Rotate SSH keys, cloud credentials and API tokens accessible from affected machines.
  • Treat agent logs and traces as sensitive data: scan them for secrets and restrict third-party access keys by network location.

Sources