Skip to content
Trending
AI2 min read

Fake Claude Code Extension Hit 5,000 Cursor Users With Backdoor

Anthropic says its own systems were not breached after a tampered Claude Code extension carrying a hidden SSH backdoor spent 11 hours on Cursor's marketplace.

By The Prompt

The AI Desk · (4 hours ago)

Reported fromoutlets, credited2
Illustration of a code editor window with a warning symbol over a VS Code extension icon
Illustration of a code editor window with a warning symbol over a VS Code extension iconPhoto via The Register

A tampered copy of the Claude Code extension sat live on Cursor's marketplace for roughly 11 hours on Thursday, reaching as many as 5,000 developers before anyone caught it, according to The Register. Anthropic says the breach has nothing to do with its own infrastructure.

The poisoned package, version 2.1.313 of the official Claude Code extension for Visual Studio Code, was available between 0227 and 1335 Pacific time, Cursor told researchers. Security firm Phorion first flagged the extension as carrying a persistent SSH backdoor built to run on Windows, macOS and Linux.

Once installed, the implant collected the host machine's username, configured Git email and public IP address, then tried to open a reverse SSH tunnel back to an attacker-controlled server. There's a catch that limited the damage. Opening that tunnel required admin rights on macOS and Windows, and in most cases the script didn't have them, so the connection attempt simply failed.

That doesn't mean affected machines are clean. Even when the tunnel never opened, the extension still sent host and user data to the attacker, dropped the attacker's SSH key on the machine and installed a persistence mechanism. Cursor and Phorion both say developers who had the compromised version installed need to actively hunt for the infection, because updating to a newer extension release does not remove it on its own.

Cursor has published remediation instructions and download scripts meant to wipe the infection, laid out in its own incident writeup.

Anthropic pushes back on blame

Online chatter following the discovery pointed fingers at Anthropic, with some cybersecurity voices suggesting the tampered extension traced back to a compromise of Anthropic's own systems. Felix Rieseberg, a member of Anthropic's technical staff, pushed back on that directly in a post Thursday.

"We're working with Cursor on this incident, but it wasn't a compromise of Anthropic's systems or employee laptops," Rieseberg wrote. "A malicious actor used access to Cursor's marketplace to publish a tampered copy of the Claude Code extension, which we didn't build or publish. The versions on our official channels are not affected, and this only impacts Cursor users."

Cursor's own account backs that framing without explaining how the attacker got into its marketplace in the first place. "We removed the malicious version from the marketplace immediately upon detection and blocked further distribution," the company said in its advisory, adding that it shut down the entire VS Code Marketplace as a precaution, checked other extensions for similar tampering, rotated its keys and worked with a cloud provider to cut off the attacker's server.

Neither Cursor nor Anthropic has said publicly how the attacker gained the access needed to push a malicious update to begin with. Anyone who installed the Claude Code extension through Cursor's marketplace around that 11-hour window on Thursday should check Cursor's advisory and run its cleanup scripts rather than assume a fresh install has fixed things.

Claude Code: everything we knowPrice, specs, release date and deals in one place

Reported from

How this story was made

Written by the Hitechreports desk from the reporting credited above, with facts attributed to their original publishers. We do not test devices ourselves; anything about performance, battery life or cameras comes from the outlets that did. Prices are as reported at the time of writing. Editorial policy · Report an error

What did you make of it?

Filed by The AI Desk

Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.

More from this desk →

Be the first to comment

Join the argument. No password, just your email or a passkey.

    Read next

    More AI →