Act nowSupply chain

A fake Claude Code extension 2.1.313 sat in the Cursor marketplace with an SSH backdoor: check your keys

Cursor said on 29 Oct 2026 that a tampered build of the Claude Code extension, version 2.1.313, was in its marketplace for about 11 hours and tried to plant an SSH backdoor. At most 5,000 users. Here is how to check and clean up.

Updated 15 Nov 2026: see what changed

How urgent

Act now

Who is affected

Anyone who installed or updated the Claude Code extension (Anthropic.claude-code) to version 2.1.313 from the Cursor marketplace on 29 October 2026, between about 2:27am and 1:35pm Pacific time. Cursor says at most 5,000 users

What to do

Check whether you got version 2.1.313 from Cursor on 29 October and look for the key named svc-telemetry in your SSH authorized keys file. If it is there, treat the machine as compromised: remove the key and the persistence, and rotate every credential reachable from it

Published 30 Oct 2026 · Sources last checked 15 Nov 2026No. 26 / 87

Updates, follow-ups and corrections

  1. Update · 15 Nov 2026. Cursor published a full update on 13 Nov 2026 that explains how the tampered build got into its marketplace. Cursor says an attacker used an access key it had issued to Fireworks AI, which Fireworks exposed, to copy data from a read-only storage bucket the two shared. The bucket held mostly 2025 coding-agent traces from sessions that did not use Privacy Mode, and some contained unredacted credentials. Cursor says a credential taken from that exposed data let the attacker publish the tampered extension on 29 October, and that the tampered versions were 2.1.312 and 2.1.313. It says at most 5,000 users could have received the build, and that it was served only from its own marketplace, never from Open VSX. Cursor also says fewer than 0.05% of users outside Privacy Mode appeared in the traces and each was notified directly, API credentials of fewer than 0.01% of customers were involved, and Privacy Mode data, production systems, the app and its updates were not affected. The last attacker activity it saw was on 1 November. What to do is unchanged: check for the svc-telemetry key, clean up, and rotate what the machine could reach; if Cursor sent you a notice, follow it. Fireworks AI's own bulletins (read 15 Nov) do not mention Cursor or a shared bucket, so the link between the two incidents rests on Cursor's account alone.
  2. Update · 2 Nov 2026. Sophos X-Ops, citing analysis by Phorion, independently describes the same tampered build and the same behavior: an attacker's SSH key, a reverse tunnel attempt and persistence. Sophos says SSH could not be enabled on most macOS and Windows machines because of coding errors, and that removing the extension does not remove the backdoor: the key, the persistence and any tunnel processes must be removed by hand and reachable credentials rotated. Sophos lists the server address 134.209.82[.]225 as an indicator; it gives no key name, so keep checking for the svc-telemetry key from Cursor's notice. Sophos does not say whether any machine actually opened the tunnel.

We keep the original text. When something changes, when we follow a story, or when we get something wrong, we add a dated note here instead of editing silently.

What happened

On 29 October 2026 Cursor published a security notice: a modified version of the Claude Code extension (Anthropic.claude-code, version 2.1.313) was available in the Cursor extension marketplace. Cursor states that this version was not developed by Anthropic. It was online for about 11 hours, from roughly 2:27am to 1:35pm Pacific time, and Cursor says at most 5,000 users were affected. Cursor removed it, blocked further distribution, shut the marketplace down as a precaution, rotated its keys, rebuilt it and checked that no other extension had been tampered with.

What the malicious code did

According to Cursor's notice, the code:

  • Collected host details: hostname, username, operating system, home folder, names of environment variables, Git email and public IP.
  • Tried to open SSH access: it added an attacker's key to the machine's SSH authorized keys file, tried to switch on the local SSH service and tried to open a reverse SSH tunnel. Cursor says switching on SSH usually fails without admin rights, but the key and the persistence were still installed on most machines.
  • Persisted on the machine: a LaunchAgent on macOS, a scheduled task on Windows, with scripts that retried the tunnel about every 60 seconds.
  • Reported back to the attacker's server.

Cursor says the attacker's server is now offline, so the tunnel and the reports no longer work. It still asks affected users to clean up, and it warns that uninstalling the extension may not be enough, because the persistence scripts keep running on their own.

Who is affected

Only people who installed or updated the Claude Code extension from the Cursor marketplace on 29 October within that window. Cursor says it will contact affected users directly. Cursor reports no evidence that other versions of the extension or the official distribution were affected. If you use Claude Code from the terminal, or you did not install or update the extension in Cursor that day, this story does not apply to you.

What to do

Steps follow Cursor's notice; the numbering is ours.

  1. Check the version. In Cursor, open the extensions list and see whether Claude Code is at 2.1.313 and when you installed it.
  2. Look for the key. Search the SSH authorized keys file of your user for an entry labelled svc-telemetry. On Windows, also check the shared administrators file that OpenSSH uses. Cursor's notice gives the exact command for macOS and the exact script for Windows. If the key is there, the machine is affected.
  3. Isolate and clean. Disconnect the machine, then remove the key and the persistence. Cursor provides cleanup scripts in its notice, one for macOS and one for Windows. Read any script before you run it and download it only from Cursor's own page. Decide whether to rebuild the machine.
  4. Rotate what the machine could reach, from a different, trusted device: SSH keys, tokens, API keys and passwords. Cursor notes that the data it collected was limited, but that does not rule out later access.

If you build with AI tools

Our advice, not from the notice: an AI coding extension runs with your permissions, so a tampered update can read your keys and open a door into your machine. Keep production keys out of the machine you code on, give each tool its own limited key, and after any security notice about a tool you use, check the version and your SSH authorized keys the same day.

What we could not confirm

Cursor's 13 November update explains how the attacker got a tampered version into its marketplace, but it does not name the credential used to publish it, and it does not say who they are. Fireworks AI's bulletins do not mention Cursor, so the key story rests on Cursor's account alone. Its notice covers only the Cursor marketplace; it says Open VSX never had the malwareMalware Any software made to do harm, such as stealing data, spying, locking files or giving an attacker control of a device. Tap again to open the glossary, and one earlier press report suggested the same package appeared elsewhere, which we could not confirm. The count of "at most 5,000" is Cursor's own upper limit, not a confirmed number of infected machines.

Watch the short video

The video loads from YouTube (privacy-enhanced mode) only when you press play.

Comments

No account needed. Add a nickname if you like, or leave it empty to comment anonymously. Comments are reviewed before they appear. Links are not allowed.

    We store your comment, your nickname and a scrambled form of your connection that we use only to limit spam. We do not store your IP address or set cookies. Cloudflare Turnstile checks that you are not a bot.

    Sources and verification

    We use primary sources first and say when we rely on a news report. Dates are the day we last checked each source.

    Spotted a mistake? Email [email protected] and we will correct it and note the change.

    Get the free runbooks

    One-page security checklists for founders who build with AI. You get a confirmation email first, then the runbooks.

    Send me the runbooks