November 19, 2026
The Tampered Claude Code Extension in Cursor’s Marketplace: What It Did and How to Check
For about eleven hours on 29 October, a modified Claude Code extension planted an SSH backdoor. How it worked and how to clean up.

By Shafeequ Ahmed
4 min read
I used AI tools to help draft this story from my SimpleTech video and the sources behind it.
On 29 October 2026, Cursor published a security notice about its VS Code extension marketplace. Modified builds of the extension Anthropic.claude-code, versions 2.1.312 and 2.1.313 for Windows and macOS, had been available there between about 2:27am and 1:35pm Pacific time that day. Cursor states that these malicious versions were not developed by Anthropic.
If installed, the extension tried to plant an attacker's SSH key, open a tunnel back to the attacker's server, make itself persistent across reboots, and send a report about the machine. The security company Phorion published a technical advisory the following day.
This article explains the mechanism at a level that helps you assess your own exposure and avoid the next incident, based on those two sources. It does not cover how the package was tampered with, because neither source says. When I checked on 1 November, I did not find a separate Anthropic statement, so I quote only Cursor and Phorion.
Timeline and response
According to Cursor's notice, the company removed the affected versions as soon as they were detected, then shut down the whole marketplace as a precaution. It audited access and found no other extensions tampered with, rotated its keys, rebuilt the marketplace and brought it back online. It also worked with the hosting provider to take the attacker's server offline, so the tunnel and the reporting endpoint no longer function.
Cursor says it has no evidence that other versions of the extension, or the official distribution, are affected. Phorion's advisory, published on 30 October, was later revised to add the second version number after Cursor updated its notice.
Why an extension can do so much
Phorion puts the underlying issue simply: extensions run with the permissions of the IDE and the user running it, so a backdoored extension gets immediate access to local files, source code and developer tooling as soon as its module is loaded.
Most developers picture an extension as something that waits for a command. In practice, any code at the top level of an extension's main file runs when the editor loads that file, before the extension's activate function is called. Phorion found the malicious code prepended to the start of the extension bundle, with the rest looking like the normal extension. No click was needed.
// extension.js (illustration of load order only)
console.log("top level: runs when the file is loaded");
function activate(context) {
console.log("activate: runs later, when the editor activates it");
}
module.exports = { activate };// extension.js (illustration of load order only)
console.log("top level: runs when the file is loaded");
function activate(context) {
console.log("activate: runs later, when the editor activates it");
}
module.exports = { activate };For the video I wrote a harmless extension that only records this order, and its test confirms that the top-level line runs before activate.
What the payload attempted
Both sources describe the same behaviour:
- SSH key. It added an attacker-controlled public key, labelled
svc-telemetry, to the user's SSH authorised keys. Anyone with the matching private key could log in as that user, if SSH was running. - SSH service. It tried to enable the machine's SSH service.
- Reverse tunnel. It opened an outbound SSH tunnel to the attacker's server, so the attacker could reach back in. Background scripts retried roughly every 60 seconds.
- Persistence. On macOS, a LaunchAgent named to look like an Apple service (
com.apple.sysmond). On Windows, a scheduled task calledSysMon, with a Startup-folder script as a fallback.
It also sent a one-time report including hostname, username, operating system, home directory, Git email, public IP and the names of environment variables. Phorion notes it collected names, not values, which still gives an attacker a quick inventory of which cloud credentials or API tokens live on the machine.
Why it partly failed, and why that is not reassuring
Enabling the SSH service needs administrator rights. Cursor says that outside highly unusual circumstances, such as Cursor being started with administrator privileges, this step would fail because of errors in the malicious code. Phorion identified one such error on macOS: without admin rights, the system's remote-login query returns an error message rather than a status, the script's check did not match, and the step was skipped. Cursor concludes that on those hosts the attacker could not have logged in through the tunnel.
The rest still happened. Cursor states that the extension added the attacker's key, installed persistence and sent host details. Phorion adds that if remote login was already enabled, or turned on later, the backdoor would be ready. Cursor's advice is to follow the remediation steps regardless.
How to check your machine
Cursor's notice includes a one-line check for macOS that looks for the planted key:
grep -qs svc-telemetry ~/.ssh/authorized_keys && echo "svc-telemetry key found" || echo "No key found; machine not impacted"grep -qs svc-telemetry ~/.ssh/authorized_keys && echo "svc-telemetry key found" || echo "No key found; machine not impacted"The notice has an equivalent PowerShell command for Windows, which also checks the system-wide administrators' authorised keys file. If the key is found, the machine is impacted. A clean result means only that this published indicator is absent; Cursor also lists further indicators for endpoint and network telemetry.
If you were affected, Cursor's steps are:
- Isolate the machine.
- Remove the unauthorised key, the persistence mechanisms and any running tunnel processes, and assess whether to rebuild. Cursor provides remediation scripts for macOS and Windows that exit with a non-zero status if anything remains.
- Do not rely on uninstalling. Cursor says it may not be sufficient because the payload drops standalone scripts that keep running; Phorion says uninstalling alone will not remove the backdoor.
- Rotate credentials and sessions the affected account could reach, from a trusted system. Cursor notes the limited collection does not rule out later credential access over the backdoor.
On a work laptop, involve your security team rather than cleaning up quietly.
Habits that reduce the next blast radius
These are my recommendations rather than findings from the sources:
- A familiar name is not a guarantee. Check the publisher and which marketplace your editor actually downloads from.
- Do not run your editor as an administrator. Cursor's notice describes admin rights as the unusual case in which the SSH step could have worked.
- Keep long-lived secrets out of your everyday shell environment where you can.
- Review installed extensions periodically and remove what you no longer use. Every extension is code that runs as you.
Sources
- https://cursor.com/blog/claude-code-extension-security-notice
- https://phorion.io/blog/offical-claude-code-vsx-compromise/
I explain this in more detail, with diagrams, in my SimpleTech video: https://www.youtube.com/watch?v=iluQpJaeH4g More developer explainers on my channel: https://www.youtube.com/@SimpleTechFirstCommit