Skip to content
InYourGeek
visiteur@inyourgeek — shell
↹ compléter↑↓ historique⏎ ouvrirhelp
FR

The Monday letter · the tech and security week in one email

Free, every Monday, via Brevo (EU), one-click unsubscribe · your data and rights

Security· 3 min read

Cursor confirms a breach and a booby-trapped Claude Code extension

Cursor says a key stolen from its vendor Fireworks AI let attackers push rigged copies of the Claude Code extension to its marketplace on October 29. Up to 5,000 developers may have received code that installs an SSH backdoor.

Illustration of a code editor's extension marketplace with a tampered Claude Code extension package and an SSH key slipping out of it

Code editor Cursor published an incident report on Friday, November 13, 2026 (yes, Friday the 13th, make of that what you will) confirming a breach of its systems. The story comes from Bleeping Computer in a piece dated November 14, which is our only source: what follows is Cursor’s own account, as that outlet summarises it.

A key left at the vendor’s

According to Cursor, it all started with an access key held at Fireworks AI, a third-party vendor used for model training and inference. That key reportedly let the attackers copy the contents of a read-only storage bucket Cursor used to share data with Fireworks.

The bucket held historical coding-agent traces, most of them dating from 2025. They came from users who had not switched on Privacy Mode and whose data could be used for training. Some of those traces contained unredacted API credentials, which are said to have opened the door to other Cursor systems. Read-only, as it turns out, is plenty when the thing being read is a set of keys.

The company says fewer than 0.05% of users had traces exposed, and fewer than 0.01% of customers had API credentials compromised.

What reportedly walked out the door

Armed with those credentials, the attackers are said to have pulled data from several SaaS tools Cursor uses. The list published by the company includes:

  • archived, non-production source code repositories, some of which contained business contact details;
  • support information and email exchanges, in limited quantities;
  • usage metrics and customer contact details;
  • summaries of some conversations with agents;
  • credentials that customers had sent to support.

Cursor says its monitoring spotted the activity, which led it to revoke the Fireworks key and rotate the exposed credentials. One problem: by then the attackers already held a credential that allowed publishing to the extension marketplace. The locks were changed, minus one.

Eleven hours on the shelf on October 29

On October 29, unauthorised copies of Anthropic’s Claude Code extension were reportedly dropped straight into the cache that feeds Cursor’s extension marketplace. The versions in question are 2.1.312 and 2.1.313, which stayed available for roughly eleven hours before being pulled. Neither Anthropic’s legitimate extension nor Open VSX, the marketplace Cursor builds on, was compromised.

Per Bleeping Computer, the added code collected system information, installed an SSH key controlled by the attackers, tried to enable SSH access and set up persistence mechanisms. Security researchers found coding errors in it that would probably have stopped the backdoor from working on many Windows and macOS machines where SSH was not already enabled. Even malware ships with bugs, though that is not a patch you should count on.

Cursor estimates that up to 5,000 users who requested those versions during the exposure window may have received the rigged ones.

What you need to check

The first question is simple: did you install or update the Claude Code extension in Cursor on October 29, and did you get version 2.1.312 or 2.1.313? If so, Cursor points you to its remediation guide, the contents of which our source does not detail. Given the behaviour described, the signs to look for are an SSH key you did not add, or SSH access that switched itself on without your help.

According to the company, production infrastructure, official releases of the app, customer code repositories, and the code and prompts of Privacy Mode users were not affected. Cursor says it has pulled all its inference traffic from Fireworks, rebuilt its marketplace and redeployed its production infrastructure as a precaution.

One lesson travels well beyond Cursor: credentials left lying in agent traces in 2025 became the foothold for a breach in 2026. If you have ever pasted a secret into a chat with an assistant or a support ticket, it is no longer a secret, just a rotation you have not done yet.

Sources (1)

Written with AI assistance from the sources cited above, then reviewed and approved before publication by Sébastien Soulier.

The Monday letter

The tech and security week, every Monday in your inbox

What to patch first, data breaches, the articles with pictures. Free, no ads, one-click unsubscribe.

Free, every Monday, via Brevo (EU), one-click unsubscribe · your data and rights