Trojanized Claude Code Extension on Cursor: What We Know

Dileep Solanki


Trojanized Claude Code Extension Reached Cursor Users: What We Know So Far

By Dileep solanki · Published: 1 November 2026 · This is a developing story. Details may change as investigations continue.

On 29 October 2026, a tampered copy of the Claude Code extension appeared in Cursor's extension marketplace. It carried code that tried to give an attacker remote access to developers' machines. Within a day, headlines were claiming that Anthropic had been "hacked" and that laptops worldwide were exposed. The documented facts are narrower, and the details matter if you need to decide whether to act.

What happened

According to security firm StepSecurity, versions 2.1.312 and 2.1.313 of the Claude Code extension (anthropic.claude-code) were published to Cursor's marketplace for macOS and Windows, carrying a malicious loader inside the extension's extension.js bundle. The malicious build ran for roughly 11 hours. Cursor reported it was available between about 2:27 AM and 1:35 PM Pacific time.

The tampered version appeared under the real publisher name, not a look-alike. It used the extension's legitimate publisher identity, and how it entered the registry has not been publicly established.

Who was affected, and how many

Cursor's figure is at most 5,000 users. That is a ceiling for people who received the build, not a count of confirmed compromises.

The "tens of millions" figure in the original post describes the extension's popularity. StepSecurity notes it has tens of millions of lifetime downloads on Open-VSX alone, but that is not the number of people exposed. Anthropic says the versions on its official channels are not affected, and only Cursor users were impacted. Microsoft likewise confirmed that its Visual Studio Marketplace and the Claude Code npm distribution were unaffected.

What the malicious code did

Per StepSecurity's sandbox and registry analysis, on load the extension planted an attacker SSH key, installed per-OS persistence, opened a reverse SSH tunnel to attacker infrastructure, and sent host reconnaissance data over plain HTTP. Notably, no Claude conversation was needed. Loading the extension was enough.

Reporting based on Microsoft's analysis adds two limits. The loader collected the names of environment variables that look like credentials, not their values, so this collection does not on its own show that the underlying secrets were stolen. And attempted execution does not prove successful intrusion: Cursor reported that errors and administrator-permission requirements blocked SSH access on most macOS and Windows hosts.

Who is responsible? The dispute

Anthropic's position, relayed by StepSecurity, is that this was not a compromise of Anthropic's systems or employee laptops. A malicious actor used access to Cursor's marketplace to publish a tampered copy that Anthropic did not build or publish.

Security researcher Kevin Beaumont, who first raised the alarm, challenged the wording of Cursor's statement. He argued that "developed" does a lot of work, because the build was uploaded to Anthropic's verified account and Cursor automatically updates extensions. Both points can be true at once: Anthropic's own infrastructure was not breached, yet users saw a trusted publisher name on a malicious update. As of writing, the attacker's identity and the exact way they gained marketplace access have not been publicly established.

One more caution: some early reports said the malicious build came through Open-VSX. StepSecurity later corrected this, stating that Anthropic confirmed 2.1.313 was never published to Open-VSX, the Microsoft Visual Studio Marketplace, or any official host. Some third-party tools and pages still describe the older version of the story.

How to check whether you are affected

You are only in scope if you used Cursor and the extension was installed or auto-updated from Cursor's marketplace during the window. StepSecurity's guidance is to treat a machine as compromised if anthropic.claude-code 2.1.313, or 2.1.312 installed from Cursor's marketplace, was present between October 29, 09:27 UTC and October 30.

Steps from StepSecurity's and Cursor's guidance (I have not run these myself):

  1. Look for the implant key. On macOS or Linux, search your ~/.ssh/authorized_keys file for the marker svc-telemetry. On Windows, inspect %USERPROFILE%\.ssh\authorized_keys.
  2. Remove persistence, not just the extension. Uninstalling the extension alone is not sufficient, because the loader drops standalone persistence that survives removal. Cursor publishes remediation scripts for macOS and Windows.
  3. Update. Install version 2.1.314 or later.
  4. Rotate credentials reachable from that machine, such as source-control tokens, cloud keys, SSH keys, and CI tokens, working from a known-clean device.
  5. Consider a rebuild if you find evidence the tunnel actually connected.

The command-and-control server has been dealt with: Cursor worked with the hosting provider to take the attacker's server offline. Key indicator values and file paths are listed in StepSecurity's full write-up, linked below.

What this means for developers

Extensions run with your user's privileges and often update silently. This incident shows that a trusted publisher name does not guarantee the file you receive is clean when a marketplace itself is abused. Reasonable precautions include turning off automatic extension updates on machines that hold production credentials, pinning versions in managed environments, and keeping secrets out of environment variables on developer laptops. Security teams may also want allow-lists for approved publishers, which Microsoft-based guidance reported by GBHackers also suggests.

What we still don't know

  • How the attacker obtained marketplace publishing access.
  • Whether any machine was successfully accessed through the tunnel.
  • Whether the final affected-user count differs from Cursor's 5,000 ceiling.
  • Who is behind the attack.

This article will need updating as Cursor, Anthropic, and independent researchers publish more.


How this was made: I researched this piece with AI assistance, using the sources below, and edited it against them. I could not open Cursor's own advisory or Microsoft's blog post directly, so statements attributed to them come from reporting by StepSecurity and GBHackers. I did not test any malware, command, or remediation step. This is general information, not incident-response advice for your organisation. If you suspect compromise, contact your security team.

G. Sources and Further Reading

ALSO VIST OUR WEBISTES FOR NEWS AND AI RELATED ARTICLES 

THINKTODAYNEWS :- 
https://thinktodaynews.media/

AIBHARAT.ME :-  https://www.aibharat.me/
3/related/default