Technical Lead / Web, SaaS & AI
Home / Blog

Claude Code Extension Backdoor 2.1.313: How to Check

Get new posts by email

Security alerts, SEO and web development guides, sent when they are published.
No spam. Unsubscribe anytime. Privacy Policy
Claude Code extension backdoor: a developer laptop at night showing an unusual activity warning from an unknown IP address

Updated: November 11, 2026. This is a developing story, and we will update this guide as new details are confirmed.

Latest update, October 31: Cursor has confirmed the compromise happened in its own marketplace, and says the attacker could not log in through the tunnel on machines where SSH could not be switched on. Cleanup is still required, and how the attacker got access is still unknown. See what is confirmed and what is not.

For about 11 hours on October 29, 2026, a tampered copy of the Claude Code extension, version 2.1.313, was served to developers through Cursor's extension marketplace. It looked and worked like the real thing, but hidden inside was code that tried to plant an SSH backdoor on the developer's computer. This Claude Code extension backdoor spread quietly through auto-updates, which is exactly what makes supply chain attacks so dangerous.

Cursor says at most 5,000 users were affected. You may have seen much bigger numbers on social media, such as "29 million installs". That figure is the install count of a different marketplace listing, not the number of victims.

In this guide, you will learn what happened, what is fact and what is still unclear, and exactly how to check and clean your machine on macOS, Windows and Linux.

Used Claude Code in Cursor on October 29? Skip straight to How to Check Your Machine and run the checks now.

Claude Code Extension Backdoor at a Glance

Detail Information
Affected package Anthropic.claude-code version 2.1.313
Where Cursor's own VS Code extension marketplace, confirmed by Cursor
When About 2:27 am to 1:35 pm PT, October 29, 2026
Users affected At most 5,000, according to Cursor
What it did Tried to install a persistent SSH backdoor and send host details to the attacker
Platforms macOS and Windows builds confirmed by Cursor; Linux persistence via crontab also reported
Clean version 2.1.314, released October 30 on Open VSX, the VS Code Marketplace and npm
First detected by Phorion, a macOS security firm
Attacker server Now offline, according to Cursor

What Happened: The 11-Hour Timeline

Here is how the incident unfolded, based on public statements and security research.

  • October 29, about 2:27 am PT. A malicious version 2.1.313 appears in Cursor's VS Code extension marketplace. It is listed under Anthropic's real extension ID, anthropic.claude-code, rather than as a lookalike. Anyone with auto-update turned on could receive it without doing anything.
  • October 29. Phorion detects the backdoor and publishes an advisory. Its first report said the package came from Open VSX, the open extension registry used by VSCodium and other editors. Later reporting corrected this to Cursor's own marketplace.
  • October 29, about 1:35 pm PT. Cursor removes the malicious version and blocks further distribution. As a precaution, it shuts down its entire VS Code marketplace, audits access, confirms no other extensions were tampered with, rotates its keys, and restores the marketplace.
  • October 29. Cursor publishes a security notice with indicators and cleanup scripts, and says it will contact affected users directly. It also works with the hosting provider to take the attacker's server offline.
  • October 29 to 30. Security researcher Kevin Beaumont publishes a post headlined "Anthropic hacked...", which spreads quickly on social media.
  • October 30. Anthropic responds that this was not a compromise of its own systems. A clean version, 2.1.314, is published to Open VSX, the Visual Studio Marketplace and npm within about an hour of each other.
  • October 30 to 31. Cursor tells researchers at OpenSourceMalware, in its updated report, that the extension in its marketplace "experienced some form of compromise", and that the malicious package came from its own marketplace, not Open VSX. The Hacker News adds an editor's note tracing the incident to a compromise of the Cursor VS Code Marketplace. Cursor also clarifies that on most machines the attacker could not log in through the tunnel.

Was Anthropic Hacked? Facts vs Claims

This is the question most people are asking, and the answer depends on separating what has been confirmed from what has been claimed.

What Cursor says: the malicious version "was not developed by Anthropic". Cursor says it has no evidence that other versions of the extension or the official distribution were affected, and that no other extensions in its marketplace were tampered with.

What Anthropic says: in Anthropic's response to the compromise claims, Felix Rieseberg of Anthropic's technical staff said it was not a compromise of Anthropic's systems or employee laptops. Instead, a malicious actor used access to Cursor's marketplace to publish a tampered copy of the extension, which Anthropic did not build or publish. He added that Anthropic's official channels were not affected.

What critics say: Kevin Beaumont argues that Cursor's wording sidesteps an important point. The package appeared under Anthropic's real extension ID and publisher name, not as an obvious fake. Security reporters have made the same observation.

What Cursor has confirmed since: Cursor has confirmed that the compromise happened in its own marketplace, telling researchers the extension there "experienced some form of compromise". It has not yet said how the attacker gained that access.

One conflicting report: StepSecurity's analysis says 2.1.313 was published to Open VSX first and reached Cursor through Cursor's mirror of it. However, Open VSX, the VS Code Marketplace and npm all jump straight from version 2.1.312 to 2.1.314, with no 2.1.313 listed, and most sources, including Cursor, point to Cursor's own marketplace.

What is still unknown:

  • How the attacker got publish access to Cursor's marketplace under Anthropic's name.
  • Whether any affected machine was actually accessed through the backdoor.
  • Linux-specific indicators from Cursor, whose notice covers macOS and Windows only.

The bottom line: based on the public evidence so far, this was a supply chain compromise of the extension as distributed through Cursor's marketplace. According to OpenSourceMalware, as reported by Rescana, the npm package, the Open VSX extension and the Microsoft Marketplace extension are not malicious. Headlines saying "Anthropic was hacked" go further than the evidence currently shows. If new facts emerge about how the package was published, we will update this section.

How the Claude Code Extension Backdoor Worked

The attack was simple, but well disguised.

A small loader hidden inside a real extension

According to The Hacker News analysis of the trojanized extension, the package was the genuine Claude Code extension with a malicious 66-line loader inserted at the top of its main JavaScript file. Everything else looked normal, so the extension worked as expected and gave users no reason to suspect anything.

Claude Code extension backdoor loader: obfuscated JavaScript inserted at the top of the extension main file

It ran as soon as the extension loaded

The malicious code ran when the extension loaded in the editor. You did not need to open Claude Code or type a prompt for it to start.

What it did, step by step

  1. Collected information about your machine: hostname, username, operating system, processor architecture, home and working directories, your Git email, your public IP address, and the names (not the values) of up to 40 environment variables containing words like KEY, TOKEN, SECRET, PASS, AWS or GCP.
  2. Added an attacker's SSH key, labeled svc-telemetry, to your authorized_keys file. This is the key that would let the attacker log in.
  3. Tried to turn on the SSH service on your computer so it could accept remote logins.
  4. Tried to open a reverse SSH tunnel to the attacker's server, which lets the attacker reach your machine even from behind a router or firewall.
  5. Set up persistence so it would survive restarts: a LaunchAgent on macOS, a scheduled task or Startup script on Windows, and a crontab entry on Linux. Background scripts retried the tunnel about every 60 seconds.
  6. Reported back to the attacker by sending your machine details over HTTP to port 8099.

Why a failed step is not a reason to relax

Cursor notes that turning on the SSH service normally needs administrator rights, so that step usually failed on macOS and Windows unless the editor was running as an administrator. That is good news, but it is not the end of the story.

According to Help Net Security's report on Cursor's clarification, Cursor says that on machines where SSH could not be enabled, the attacker could not log in through the tunnel. Still, the key, the persistence scripts and the data sent to the attacker remain, so cleanup is still required.

Are You Affected?

Use this table to see where you stand.

Where you installed Claude Code Affected?
Cursor, and the extension installed or updated to 2.1.313 between 2:27 am and 1:35 pm PT on October 29 Possibly. Check your machine now.
Cursor, but the extension was not updated during that window Very unlikely, but the quick check takes one minute
VS Code, from Microsoft's Visual Studio Marketplace No. Listed as not malicious, and no 2.1.313 was published there
Open VSX (VSCodium and other editors) No, according to Cursor and OpenSourceMalware; one report disagrees (see above)
Claude Code command-line tool (installed with npm) No. No version 2.1.313 was published there

The Hacker News reports that the malicious update was distributed only through Cursor's marketplace, with no evidence it reached Open VSX or Microsoft's VS Code Marketplace. Anthropic also says its official channels were not affected.

The biggest risk factor was auto-update. If Cursor updated your extensions automatically on October 29, you could have received the malicious version without clicking anything.

How to Check Your Machine

The key test is simple: look for an SSH key labeled svc-telemetry. If you find it, treat the machine as compromised and move on to the cleanup section. The other checks help confirm what else was installed.

Checking for the Claude Code extension backdoor: a terminal log warning of outbound activity from svc-telemetry

macOS

Open Terminal and run these checks:

  1. Look for the attacker's SSH key:
grep -n "svc-telemetry" ~/.ssh/authorized_keys
  1. Look for the persistence files:
ls -la ~/Library/LaunchAgents/com.apple.sysmond.plist ~/.config/.smt ~/.config/.smtk 2>/dev/null
launchctl list | grep sysmond
  1. Check whether the bad version is installed in Cursor:
ls ~/.cursor/extensions | grep claude-code

If any of these return a result that mentions svc-telemetry, com.apple.sysmond, .smt or claude-code-2.1.313, your machine was affected.

Windows

Open PowerShell and run these checks:

  1. Look for the attacker's SSH key in both possible locations:
Select-String -Path "$env:USERPROFILE\.ssh\authorized_keys","C:\ProgramData\ssh\administrators_authorized_keys" -Pattern "svc-telemetry" -ErrorAction SilentlyContinue
  1. Look for the scheduled task and Startup script:
Get-ScheduledTask -TaskName "SysMon" -ErrorAction SilentlyContinue
Test-Path "$env:APPDATA\Microsoft\Windows\Start Menu\Programs\Startup\sysmon.cmd"
Test-Path "$env:USERPROFILE\.config\smt.cmd"
  1. Check whether the bad version is installed in Cursor:
Get-ChildItem "$env:USERPROFILE\.cursor\extensions" | Where-Object Name -like "*claude-code*"

A result for the key, a task named SysMon, or a True from either Test-Path line means the machine was affected.

Linux

Cursor's notice covers macOS and Windows, but researchers report that the payload also used crontab for persistence on Linux. Check carefully:

grep -n "svc-telemetry" ~/.ssh/authorized_keys
crontab -l
ls ~/.cursor/extensions | grep claude-code

In the crontab -l output, look for any entry you do not recognize, especially ones that start SSH or run scripts from hidden folders such as ~/.config.

Check for an active tunnel

On any system, look for running SSH tunnel processes:

ps aux | grep -E "autossh|ssh -N -R" | grep -v grep

On Windows, use:

Get-CimInstance Win32_Process | Where-Object { $_.CommandLine -match "autossh|ssh -N -R" } | Select-Object ProcessId, CommandLine

If your company has network logs, also look for connections to 134.209.82.225 on ports 22 or 8099.

How to Clean Up: Why Uninstalling Is Not Enough

It is natural to think that removing the extension removes the problem. It does not. The malware installs standalone persistence scripts that keep running on their own, even after the extension is gone. Updating to version 2.1.314 does not remove them either.

Isolating a compromised developer laptop by unplugging the network cable before backdoor cleanup

If your checks found anything, follow these steps:

  1. Isolate the machine. Disconnect it from the network, or at least from company VPNs, until cleanup is done.
  2. Run Cursor's official cleanup scripts. Cursor's official security notice and remediation scripts include one script for macOS and one for Windows. On macOS, run the script as the affected user, not as root. On Windows, run it from an elevated PowerShell window (Run as Administrator). Always download the scripts directly from Cursor's notice.
  3. Confirm the result. The scripts remove the extension, the persistence items, any running tunnel processes and the SSH key. They exit with an error if anything is left behind, so read the output carefully.
  4. Clean Linux machines manually. Open ~/.ssh/authorized_keys in a text editor and delete the line that ends with svc-telemetry. Then run crontab -e and remove any entry you do not recognize. Check for and stop any running ssh -N -R or autossh processes.
  5. Re-run the checks from the previous section to make sure nothing remains.
  6. Update the extension to version 2.1.314 or later.
  7. Decide whether to rebuild. If the machine has access to production systems, customer data or signing keys, rebuilding it from a clean image is the safest option. You cannot be sure what an attacker did through an open backdoor.

What to Rotate Afterwards

The malware sent the names of your environment variables to the attacker, not their values. But names alone reveal a lot: which cloud providers, CI systems and services you use. And a working SSH backdoor could have allowed the attacker to read the values directly.

Rotating API keys and secrets after the Claude Code extension backdoor, with MFA and a hardware security key

From a trusted, clean machine, rotate:

  • Cloud provider keys (AWS, Google Cloud, Azure)
  • API tokens and secrets, especially those whose names include KEY, TOKEN, SECRET or PASS
  • GitHub, GitLab and other Git credentials and personal access tokens
  • CI/CD secrets and deployment keys
  • Your own SSH keys
  • Active sessions in tools you were signed in to on that machine

If your team needs help working through this, a developer workstation security review can map which credentials each machine could reach.

Lessons for Developers and Teams

The Claude Code extension backdoor is a reminder that the tools developers trust most are now prime targets. It is part of a wider pattern this month, from the UpdraftCentral backdoor attack on WordPress sites to the AuthMask vulnerability in Microsoft Authenticator, where a malicious app could pose as a trusted one. Attackers are also moving faster than ever, as AI-automated attacks like CyberStrikeAI show.

Development team reviewing approved editor extension versions on a shared dashboard

Here is what teams can do:

  • Treat extensions like production dependencies. They run with your full user permissions and can read your files, keys and code.
  • Delay auto-updates on critical machines. A short delay gives the community time to spot a bad release before it reaches you. VS Code already waits before installing some extension updates by default, which helped protect its users here.
  • Pin and approve extension versions for your team, and keep an allowlist of approved extensions.
  • Know your marketplace. VS Code forks such as Cursor, Windsurf and VSCodium can use different extension marketplaces, each with its own security controls.
  • Monitor for SSH key changes. An alert on any change to authorized_keys would have caught this attack immediately.
  • Build security into your workflow. Good secure software development practices cover the tools you build with, not just the code you ship.

For the full set of controls, from extensions and SSH keys to secrets and AI coding tools, see our checklist for securing developer laptops.

Indicators of Compromise (IOCs)

Security teams can use these indicators, taken from Cursor's security notice and researcher reports, to search endpoint and network logs. IP addresses are defanged.

Type Indicator
Extension anthropic.claude-code version 2.1.313
SHA-256 (macOS, darwin-arm64) fcbb327e38e61c4fec795e0d0591f52cd1014b813d20a5912104953c3a17c364
SHA-256 (Windows, win32-x64) 3de1722901165769893bef57c72594e9a54459faa23f3091b136b79966edd081
Attacker IP 134[.]209[.]82[.]225
Ports TCP 22 (reverse tunnel), TCP 8099 (/beacon)
SSH key label svc-telemetry
Unauthorized key fingerprint SHA256:0Z6R+DF64Do2nwQic8a7kkfDLNNk4IiK+2U2Rci4htk
macOS files ~/Library/LaunchAgents/com.apple.sysmond.plist, ~/.config/.smt, ~/.config/.smtk
Windows files and tasks Scheduled task SysMon, Startup\sysmon.cmd, %USERPROFILE%\.config\smt.cmd, %USERPROFILE%\.config\smtk
Linux Unfamiliar crontab entries (reported by researchers; Cursor has not published Linux indicators yet)
Command lines ssh -N -R '*:0:localhost:22', autossh -M 0, schtasks /create /tn SysMon

Note that com.apple.sysmond imitates a real Apple naming style. Do not trust a file because its name looks official.

This alert is part of our website security for small business guide, which puts every 2026 incident into a layer-by-layer plan you can follow.

Not sure if this affects you? I've recovered 50+ attacked websites. Request a security review.

Frequently Asked Questions

Was Anthropic hacked?
Anthropic says no, and Cursor says Anthropic did not develop the malicious version. Cursor has since confirmed the compromise happened in its own marketplace. How the attacker got that access has not been explained yet.
Which Claude Code extension version was malicious?
Version 2.1.313 of anthropic.claude-code, served through Cursor's extension marketplace between about 2:27 am and 1:35 pm PT on October 29, 2026.
Was the VS Code Marketplace version affected?
No. The Hacker News reports the malicious update was distributed only through Cursor's marketplace, with no evidence it reached Open VSX or Microsoft's VS Code Marketplace, and Anthropic says its official channels were not affected. According to OpenSourceMalware, as reported by Rescana, the Microsoft Marketplace extension, the Open VSX extension and the npm package are not malicious.
How many users were affected?
Cursor says at most 5,000 users and will contact them directly. The 29 million figure shared on social media is a marketplace install count, not the number of victims.
Could the attacker log in to my machine?
On most macOS and Windows machines without admin rights, no. Cursor says SSH could not be switched on there, so the attacker could not log in through the tunnel. The key, the persistence scripts and the data already sent remain, so cleanup is still required.
How do I know if my machine was compromised?
Search your SSH authorized_keys file for an entry labeled svc-telemetry. If it is there, treat the machine as compromised.
Does uninstalling the extension remove the backdoor?
No. The malware installs standalone persistence scripts that keep running. Use Cursor's cleanup scripts or remove each item manually.
Is version 2.1.314 safe?
Yes, 2.1.314 is the clean release. But updating does not remove a backdoor that is already installed, so check and clean your machine first.
Is the Claude Code command-line tool affected?
No. Reports found no malicious version 2.1.313 published to npm for the Claude Code CLI.

Conclusion

The Claude Code extension backdoor was short-lived and limited to Cursor's marketplace, but it was well hidden and it does not clean itself up. If you used Claude Code in Cursor on October 29, check for the svc-telemetry key, run the official cleanup, and rotate your credentials from a clean machine.

For more practical security breakdowns, read more cybersecurity guides on the blog. And if your team needs a hand checking its machines, get help auditing your developer machines.

Get new posts by email
New guides and alerts, sent when they are published. No spam. Unsubscribe anytime.

Comments (0)

Related Blog

APT root flaw CVE-2026-109518 title graphic: a server rack with a package that is half verified and half tampered, linked to a hooded attacker, with the line Patch With Retries Off for Debian 12+ and Ubuntu 22.04+, fixed November 4
Security Alerts
  • 16 Nov 2026
  • <25
  • 0 Comment

APT CVE-2026-109518: Patch Your Debian and Ubuntu Servers Safely

APT CVE-2026-109518 can install a swapped package as root on Debian and Ubuntu. Patch APT with retries off, verify the version and rebuild Docker images.


Meta's 805 million euro fine: how to close Facebook's access to your photo gallery
Security Alerts
  • 15 Nov 2026
  • 356
  • 0 Comment

Meta's €805M Fine: Facebook Was Reading Your Photo Gallery. Here's How to Turn It Off

Italy fined Meta €805M because Facebook read Android photo galleries without clear consent. Turn off gallery access on Android and iPhone in 2 minutes.


WordPress update flaw CVE-2026-119977: block, then update. Exploited now, fixed in WordPress 7.1.4
Security Alerts
  • 14 Nov 2026
  • 74
  • 0 Comment

WordPress CVE-2026-119977: Why Updating Can Expose Your Site

WordPress CVE-2026-119977 is exploited, and the update to 7.1.4 itself can be raced. Block wp-content/upgrade first, then update and verify, step by step.


,