Cursor Security Incident: Malicious Claude Code Extension Versions 2.1.312 and 2.1.313 Tried to Plant an SSH Backdoor, Credentials Exposed, and What to Check This Week
Cursor confirms an attacker obtained an access key, malicious Claude Code extension versions 2.1.312 and 2.1.313 tried to install an SSH backdoor, and some customer credentials were exposed. Here is what is confirmed and what to check.
Where Worldwide
On this page 8 sections
Cursor has confirmed a security incident. An attacker obtained an access key, and malicious copies of the Claude Code extension tried to install a persistent SSH backdoor on the computers that ran them. If anyone at your business writes code in Cursor on a Windows PC or a Mac, set aside an hour this week to check. Cursor says that uninstalling the extension may not be enough.
This post keeps what Cursor has confirmed separate from what it has not said, and ends with a checklist for this week.
What happened in the Cursor security incident
Cursor published two posts: a security notice about the Claude Code extension, and a later security incident update. Together they confirm these points:
- An attacker obtained an access key that Cursor had issued to Fireworks. In Cursor's words: "An attacker obtained an access key we had issued to Fireworks."
- Malicious versions of the Claude Code extension, identified as Anthropic.claude-code, were available. Cursor says they run malicious code that tries to install a persistent SSH backdoor.
- Cursor states that these malicious versions were not developed by Anthropic.
- The notice gives a window of about 2:27am PT to 1:35pm PT. The update says the affected users requested specific cached versions of the extension on October 29.
- On October 29, Cursor said publicly that it was investigating a security incident.
Which Claude Code extension versions were malicious
Cursor names two versions:
| Extension | Malicious versions | Platforms |
|---|---|---|
| Anthropic.claude-code | 2.1.312 | Windows and macOS |
| Anthropic.claude-code | 2.1.313 | Windows and macOS |
If a computer at your company had either version installed through Cursor, assume the backdoor attempt may have run on it.
How many Cursor users were affected
Cursor gives these figures:
- Extension: at most 5,000 users requested the specific cached versions of the extension on October 29.
- API credentials: fewer than 0.01% of customers had API credentials exposed.
- Agent traces: Cursor says a limited number of coding agent traces contained credentials that had not been redacted.
Cursor also says: "The vast majority of customers were not affected, and we have contacted each affected customer directly." It adds: "We apologize to those impacted by this incident."
What is confirmed and what is not known
Confirmed by Cursor: the stolen access key, the two malicious extension versions, the SSH backdoor attempt, the time window, the upper limit of 5,000 users for the extension, the share of customers with exposed API credentials, and the unredacted credentials in some agent traces.
Not stated in the Cursor material we verified:
- How the attacker obtained the Fireworks access key.
- Exactly how that key connects to the malicious extension versions.
- How many computers actually ended up with a working backdoor. The figure of 5,000 is a ceiling on requests, not a confirmed infection count.
- Which systems the attacker may have reached through a backdoor or through the exposed credentials.
- Any specific signs of compromise to look for, or a step-by-step cleanup procedure.
If Cursor publishes more detail, follow its guidance over any general advice, including ours.
Why this matters for a company network
A developer's laptop is rarely just a laptop. It usually holds or can reach source code, cloud consoles, production servers, API keys and customer data. An SSH backdoor on that machine is a quiet way in that can outlast a password change. And because the backdoor is persistent, removing the extension that installed it does not necessarily remove the backdoor itself.
The credential exposure matters separately. API keys and tokens work wherever they are valid, whether or not the computer is ever touched again. If a key that unlocks your hosting, payment or database accounts ended up in an agent trace, the risk stays until that key is replaced.
This also applies to contractors. If an outside developer builds your website or app in Cursor, their machine is part of your attack surface.
What to do this week
Cursor has not listed a fixed extension version in the material we verified, so these steps focus on finding exposure and cleaning it up.
| Step | How to check | Why |
|---|---|---|
| Find out who uses Cursor | Ask staff developers and outside contractors, on company Windows PCs and Macs | The malicious versions were for Windows and macOS |
| Check the extension version | In Cursor's extensions list, look for Anthropic.claude-code at 2.1.312 or 2.1.313 | These are the versions Cursor names |
| Check for a message from Cursor | Search company inboxes, including developer accounts, for direct contact from Cursor | Cursor says it contacted each affected customer directly |
| Do not stop at uninstalling | If either version was installed, treat the machine as possibly compromised and have it reviewed or rebuilt | Cursor says uninstalling the extension may not be sufficient |
| Review SSH access | Look for SSH keys or remote login access nobody added, on the machine and on the servers it can reach | The malicious code tried to install a persistent SSH backdoor |
| Rotate credentials | Replace API keys, tokens and passwords used in or pasted into Cursor, starting with hosting, cloud and payment keys | Some customers had API credentials exposed, and some traces held unredacted credentials |
| Write it down | Record which machines you checked, what you found and which keys you replaced | You will need this if Cursor or a customer asks later |
If you depend on outside developers, ask them in writing whether they ran either version and what they did about it.
Cursor security incident questions people are asking
Was my company affected by the Cursor breach?
Cursor says it contacted each affected customer directly. If nobody at your company heard from Cursor, you are probably not on its list. Still check every machine for Anthropic.claude-code versions 2.1.312 and 2.1.313, because a contractor's account may not be yours.
Did Anthropic make the malicious Claude Code extension?
No. Cursor states that these malicious versions were not developed by Anthropic.
Is uninstalling the Claude Code extension enough?
Cursor says uninstalling may not be sufficient. The malicious code tried to install a persistent SSH backdoor, which can stay after the extension is gone.
How many Cursor users were affected?
At most 5,000 users requested the specific cached versions on October 29. Separately, fewer than 0.01% of customers had API credentials exposed.
Were Cursor API keys or credentials exposed?
Yes, for a limited group. Cursor says fewer than 0.01% of customers had API credentials exposed, and that some coding agent traces contained credentials that had not been redacted.
How did the attacker get in?
Cursor says an attacker obtained an access key it had issued to Fireworks. The material we verified does not explain how the key was obtained.
Which operating systems were affected?
Cursor lists the malicious versions as being for Windows and macOS.
The bigger lesson
The tools your developers use are part of your network. Extensions run with the same access as the person using them, so a tampered update can reach everything that person can reach. Keep a list of the developer tools and extensions used on company work, know who updates them, and have a plan for rotating keys quickly when a vendor reports an incident.
For help locking down developer machines and the servers they reach, see security hardening or request a quote.
Get security alerts by email, with the free checklist, at security alerts.
Sources: Cursor: Claude Code extension security notice; Cursor: Security incident update.
Have a question about this article?
Ask it here and we will answer you directly by email. Questions are private, never posted.
Security alerts by email
Hear about the next one first.
When attackers start using a flaw in software businesses run on, we email what it is, whether it touches you and what to do that week, with a round-up of our latest security writing. Free, one-click unsubscribe. You get the checklist straight away: 12 things to secure this week.
Want a result like this?
Security alerts, breaches and the technology a business runs on: what happened, who it affects and what to do.
Get a Tech QuoteKeep reading
AI Agent Attack on Epoch AI: Test Agents Tried SQL Injection on a Live Website, Trajectory Labs Pauses Testing, and What Businesses Should Check
AI agents in alignment tests run by Trajectory Labs tried to hack Epoch AI's live website, including a SQL inj...
ARTEX AI Hacking Framework: Storm-0940 Tried Sign-Ins on More Than 2,250 Tenants and Reached Privileged Access in Under 3 Hours, What Is Confirmed and What to Check
Microsoft tied a campaign to Storm-0940, which tried sign-ins on more than 2,250 tenants and in some cases rea...
Puppet Enterprise Vulnerability CVE-2026-115731: CISA Lists the Exploited Flaw, the Certificate Key Risk, and the Fixed Versions to Install
CISA added Puppet Enterprise flaw CVE-2026-115731 (CVSS 9.4) to its exploited list on November 17. Here is wha...