A backdoored Claude Code extension delivered a remote access trojan to developer laptops
A backdoored version of Anthropic's official Claude Code VS Code extension ran on developer machines on October 30, 2026. It shipped through the Cursor marketplace, auto-updated inside Cursor, and carried code that attempted to open a reverse SSH tunnel to an attacker-controlled server. Roughly 5,000 developer systems were impacted, according to Cursor.
The malicious release was version 2.1.313 of the Anthropic.claude-code package. It was pulled and replaced by version 2.1.314 within about eight hours, according to security researcher Kevin Beaumont, who published the detailed writeup. The compromised version is missing from the Cursor marketplace, the Visual Studio Marketplace and OpenVSX. Anthropic has not informed customers, the writeup says, and the bad build was deleted without notice.
Why this is a supply chain incident
The extension is the official Claude Code companion for VS Code, and tens of millions of developers use it, per Beaumont. This was not a lookalike package, and no unverified publisher was impersonating anyone. The tampered 2.1.313 was uploaded from Anthropic's own verified account.
That is the difference between a typosquat someone hides in a marketplace and a supply chain compromise: the trust chain stayed intact right up until the code changed.
What the tampered code did
Beaumont's writeup followed the first detection, made by Phorion on affected endpoints. It lists what the package attempted:
- Collect host and user information: hostname, username, OS, architecture, home directory path, environment variable names, working directories, configured Git email, and public IP.
- Add an attacker-controlled SSH key to authorized_keys, enable the local SSH service, and open a reverse SSH tunnel.
- Make the tunnel persistent through a LaunchAgent on macOS, or a Windows scheduled task that runs as SYSTEM with a startup-script fallback.
- Retry the tunnel roughly every 60 seconds from background scripts.
- Post host details and backdoor status to /beacon on port 8099 over HTTP.
He notes one important limit: unless Cursor had been started with administrator privileges, errors in the malicious code would have made the SSH step fail, and he does not read the evidence as wide remote access. He also published KQL queries for Microsoft Defender for Endpoint to hunt for the bad version, plus network indicators: SHA-256 hashes of the modified darwin-arm64 and win32-x64 VSIX packages, the reverse tunnel host 134[.]209[.]82[.]225 on port 22, and the beacon endpoint.
Whose breach is it
Cursor admitted the problem and said it identified a security incident involving a specific extension package in its marketplace. Per the writeup's update, Anthropic placed the breach on the Cursor side: "A malicious actor used access to Cursor's marketplace to publish a tampered extension targeting Cursor's users, we were not involved with building, publishing, or hosting this."
Beaumont pushes back on the phrasing. Cursor's notice describes the bad build as "not developed by Anthropic", and he calls that wordsmithing, since the version was uploaded from a verified Anthropic account.
He names Binance among notable customers that run Claude Code and Cursor on laptops, and points at a gap in the fix. Version 2.1.314 appears to remove the backdoor from the extension, but it does not remove an established SSH backdoor from systems that were already infected, nor does it unwind the prior reconnaissance.
If you use the extension, the practical move is to hunt now. Check authorized_keys, look for unexpected SSH services or LaunchAgents, and compare your VSIX package hashes against the published indicators. The window was eight hours, but the extension updated itself.
Source: doublepulsar.com
More news
Have something in mind?
Start a conversation