On 13 November 2026, the AI code editor Cursor confirmed that attackers who broke into its systems used a stolen credential to plant tampered copies of the Claude Code extension in Cursor’s extension marketplace. The fake versions tried to install a hidden SSH backdoor on developers’ computers. This explainer covers what happened, who is affected, how to check a machine in under a minute, and why it matters even if you never use Cursor.
Facts are from Cursor’s incident report and security notice, plus independent analysis, as of 15 November 2026.
The short version
- The fake builds were
Anthropic.claude-codeversions 2.1.312 and 2.1.313, for Windows and macOS, served from the Cursor marketplace only. - They were available for about 11 hours on 29 October 2026, from about 2:27 am to 1:35 pm Pacific Time.
- Cursor says at most 5,000 users could have received them.
- Anthropic’s real extension, the Visual Studio Marketplace and the Open VSX registry were not compromised.
- If you were hit, uninstalling the extension is not enough. The malware leaves its own files behind.
What happened
Cursor’s incident report describes a chain of stolen credentials:
- An attacker got hold of an access key that Cursor had issued to Fireworks AI, a company that hosts AI models. Cursor says the key was exposed by Fireworks.
- With that key the attacker copied data from a read-only storage bucket. It held coding-agent traces, mostly from 2025, from sessions where Privacy Mode was off.
- Some of those traces contained credentials that had not been redacted. The attacker used them to get into some of the software services Cursor uses internally.
- On 29 October, using one of those stolen credentials, the attacker placed malicious copies of the Claude Code extension on the cache that serves Cursor’s marketplace.
| Date (2026) | Event |
|---|---|
| 29 October | Fake builds live for about 11 hours; Cursor removes them and posts a first notice |
| 30 October | Cursor updates its notice with check scripts, adds version 2.1.312 and starts notifying affected users |
| 1 November | Last attacker activity Cursor saw, which it says was blocked |
| 13 November | Cursor publishes its full incident report |
What the fake extension did
According to Cursor’s security notice and an analysis by the security firm Phorion, malicious code was added to the start of the extension’s main file. It ran as soon as the editor loaded the extension, with no click needed. It then:
- collected details about the computer: hostname, username, operating system, Git email, public IP address and the names of environment variables that looked like keys or tokens (Phorion says the names were collected, not the values);
- added an attacker-controlled SSH key, labeled
svc-telemetry, to the user’s list of trusted keys; - tried to switch on the computer’s SSH service and open a reverse tunnel to a server the attacker controlled;
- installed scripts that retried the tunnel about every 60 seconds, even after the editor was closed.
SSH is the standard tool for logging in to a computer remotely, and the authorized_keys file lists the keys allowed to log in. Our SSH keys guide explains how that works. Adding a key to that file is like quietly giving a stranger a key to your front door.
Bugs in the malware, reported by BleepingComputer and Phorion, meant the step that switches SSH on generally failed on Macs and Windows PCs where it was not already enabled. Cursor says the attacker’s server is now offline. But the key was still added and the background scripts were still installed, so Cursor tells everyone affected to clean up regardless.
Was Claude Code itself hacked?
No. Cursor’s report says Anthropic did not publish the malicious copies and that this was not a vulnerability in Anthropic’s product. It adds that Open VSX, the registry Cursor’s marketplace pulls from, never had the malware. An Anthropic staff member said the same in a public post reported by The Register: Anthropic’s systems were not compromised, and versions from its official channels were not affected.
Anthropic’s own documentation notes that extension marketplaces are third-party software that Anthropic doesn’t maintain or audit. The copy you install is only as trustworthy as the store that serves it.
Who is affected
You are in the risk group only if all of these are true:
- you use Cursor on macOS or Windows;
- you installed or updated the Claude Code extension from Cursor’s marketplace on 29 October 2026 during the window above;
- the version you received was 2.1.312 or 2.1.313.
If you use Claude Code in a terminal, in VS Code, or in the Claude desktop app, this incident did not reach you. Cursor says its audit found no other tampered extensions in its marketplace.
How to check your machine
Cursor’s notice gives a quick check for the svc-telemetry key.
On macOS, open Terminal and run:
grep -qs svc-telemetry ~/.ssh/authorized_keys && echo "svc-telemetry key found" || echo "No key found; machine not impacted"
On Windows, open PowerShell and run:
$files = "$HOME\.ssh\authorized_keys", "C:\ProgramData\ssh\administrators_authorized_keys" | Where-Object { Test-Path $_ }
if ($files -and (Select-String svc-telemetry $files)) { "svc-telemetry key found" } else { "No key found; machine not impacted" }
Expected output on a clean machine:
No key found; machine not impacted
What to do if the key is found
Cursor’s notice sets out the response in this order:
- If you work in a team, tell your security or IT contact so they can search device and network logs for the indicators listed in the notice.
- Isolate the affected computer.
- Remove the unauthorized SSH key, the background scripts and any running tunnel processes. Cursor publishes cleanup scripts for macOS and Windows in its notice, along with the full list of files to look for. Decide whether the computer should be rebuilt from scratch.
- From a different, trusted device, change every password, API key and login session that the affected computer could reach.
The data side of the breach
The extension was one part of a wider breach. Cursor says the attacker also reached archived, non-production source code, limited customer support data and emails, product usage metrics and contact details, summaries of some agent conversations, and some credentials that users had sent to its support team.
Cursor puts the scale at fewer than 0.05% of users with agent traces exposed, and less than 0.01% of customers with API credentials exposed. It says code and prompts from Privacy Mode sessions, customer code repositories, the Cursor app and its official updates were not affected. Cursor says it has contacted affected customers. If you got a notification, rotate the keys it lists. If you did not, Cursor says your data was not in the exposed traces. Our guide on what to do after a data breach notice covers the general steps.
Why it matters
Editor extensions run with your permissions. An extension can read your files and run commands as you. Here a trusted name and a normal-looking version number were enough.
AI coding tools are now a target. The breach began with a key exposed at Fireworks AI, one of the companies in a run of October attacks on AI developer platforms that the cloud security firm Wiz calls “Black October”. Wiz says the common thread is theft of credentials, especially keys that give access to AI models.
What you paste into an agent can end up stored. The breach grew because old agent traces held real credentials.
A few habits lower the risk whatever tools you use:
- keep API keys out of chats, tickets and prompts, and rotate any key you have pasted somewhere;
- give keys the narrowest permissions the provider allows;
- turn on your editor’s privacy mode if it has one;
- look at
~/.ssh/authorized_keysnow and then, and remove entries you don’t recognize.
The same caution applies to add-ons for AI agents. If you use agent skills in Claude Code, install them only from sources you trust.
What we still don’t know
- How the key Cursor gave Fireworks was exposed. Cursor points to Fireworks’ own bulletin and gives no detail.
- Who was behind it. None of the reports we read name the attacker.
Common questions
I updated Cursor that day. Is the app itself safe? Cursor says the app, its official updates and its release signing keys were not affected.
Is it safe to install the extension now? Cursor says it rebuilt its marketplace and validated the other extensions.
Wrap-up
The Claude Code name was used as bait, but the break-in happened at Cursor, through a key held by one of its suppliers. If you used Cursor on 29 October, run the check above. If it finds the key, clean the machine and change your credentials from another device. For everyone else, the lesson is that the tools that write your code are now part of your security.
Sources
Links last checked on 15 November 2026.
- Cursor: security incident update (13 November 2026)
- Cursor: security notice on the malicious extension
- Phorion: analysis of the tampered extension
- BleepingComputer: Cursor confirms security breach
- BleepingComputer: earlier report on the SSH backdoor
- The Register: Anthropic’s response
- Wiz: Black October for AI developer platforms
- Claude Code docs: VS Code extension





Write a response