A maliciously modified version of Anthropic's Claude Code extension was available in Cursor's VS Code Extension Marketplace between about 2:27 a.m. and 1:35 p.m. Pacific on Thursday, Oct. 29. The extension ran code that attempted to install a persistent SSH backdoor. Cursor, the coding-tool company owned by SpaceX, said the incident affected up to 5,000 users.

Extensions auto-update in Cursor by default. The extension updated itself in Cursor to the malicious version.

"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," Anthropic said.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

What the code did

Cursor's notice names versions 2.1.312 and 2.1.313 for Windows and macOS and says they were not developed by Anthropic. The code collected host and user details, including the hostname, username, environment variable names, configured Git email and public IP. It attempted to add an attacker-controlled SSH key to authorized_keys, enable the local SSH service and open a reverse SSH tunnel. It attempted to make that tunnel persistent through a LaunchAgent on macOS or a scheduled task on Windows.

An analysis by Paul McCarty of OpenSourceMalware found a malicious 66-line loader inserted before the legitimate extension code. A user did not need to begin a Claude conversation. Loading the extension module was enough.

The implant gathered the names of up to 40 environment variables containing KEY, TOKEN, SECRET, PASS, AWS or GCP. Their values were not placed directly in the initial beacon. Phorion, the security firm that first noticed the activity on endpoints, said the theft of those names let the operator take inventory of cloud credentials and API tokens on a machine, likely to prioritize which machines to target in follow-up attacks.

Where the attack broke down

Cursor said a key step would have failed on most machines. Turning on the local SSH service requires admin rights, which the extension does not have on most macOS and Windows hosts. On those hosts the attacker could not have logged in through the tunnel. The attacker's server no longer functions, because Cursor worked with the cloud provider hosting it to take it offline.

Despite the coding flaws, Phorion said, "the malware is still a significant threat that installs persistence and beacons host reconnaissance back to the operator." Cursor's notice says the extension added the attacker's SSH key, installed persistence and sent host details to the attacker.

Cursor says it removed the version on detection and shut down the whole marketplace as a precaution before rebuilding it and bringing it back online. It says it audited access and confirmed no other extensions were tampered with. McCarty's analysis lists the npm package, the OpenVSX extension and the Microsoft VS Code extension for Claude as not malicious.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

What to check

Cursor's advice covers anyone who installed or updated to 2.1.312 or 2.1.313 from its marketplace on macOS or Windows during the window. The malicious extension appends the attacker's public key, labeled svc-telemetry, to the SSH authorized keys file. If that key is there, the machine is impacted.

Anthropic replaced the compromised version with 2.1.314, researcher Kevin Beaumont said. Cursor warns that uninstalling the extension may not be sufficient, because the payload drops standalone persistence scripts that keep running. It tells affected users to rotate credentials and sessions from a trusted system.

The account question

Beaumont and Phorion said version 2.1.313 was distributed through Anthropic's official account on Cursor. "It was a supply-chain compromise where the attacker gained access to the official distribution channel," Luke Roberts, Phorion's CTO and co-founder, told Ars Technica.

As of Friday, Anthropic had yet to issue public notice of the incident to customers. Neither Cursor nor Anthropic has explained how the attackers gained control of the account.

Frequently Asked Questions

Which versions of the Claude Code extension were affected?

Cursor's notice names versions 2.1.312 and 2.1.313 for Windows and macOS, distributed through Cursor's VS Code Extension Marketplace between about 2:27 a.m. and 1:35 p.m. Pacific on Oct. 29. McCarty's analysis lists the npm package, the OpenVSX extension and the Microsoft VS Code extension for Claude as not malicious.

How can I tell whether my machine was hit?

The malicious extension appends the attacker's public key, labeled svc-telemetry, to the SSH authorized keys file. If that key is there, the machine is impacted.

Does uninstalling or updating the extension remove the backdoor?

Not on its own. Cursor warns that uninstalling may not be sufficient, because the payload drops standalone persistence scripts that keep running. Cursor tells affected users to rotate credentials and sessions from a trusted system.

Could the attacker actually log in to infected machines?

Cursor said a key step would have failed on most machines, because turning on the local SSH service requires admin rights the extension lacks on most macOS and Windows hosts. The attacker's server no longer functions. Phorion still said the malware installs persistence and beacons host reconnaissance back to the operator.

How did the tampered version get published?

Beaumont and Phorion said version 2.1.313 was distributed through Anthropic's official account on Cursor. Anthropic said it was not involved with building, publishing or hosting the extension. Neither Cursor nor Anthropic has explained how the attackers gained control of the account.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

OpenAI Agents Attacked RubyGems and Tried to Steal User API Keys
OpenAI confirmed Friday that its agents used RubyGems during a May attack reconstructed from packages the attackers left in public. The agents submitted more than 2,000 packages on May 11 and 12, turn
Zed and Warp Both Ship AI Coding. They Bet on Opposite Surfaces.
One product anchors the workday to the file under your cursor. The other anchors it to the agent sessions running in the background. Zed and Warp both moved AI coding toward the center of their produ
The Files Steering Your Coding Agents Now Have a Sync Layer
San Francisco | Friday, June 5, 2026 Agent skills are hardening into their own software layer, and a Go tool named Skillshare pulled in 2,132 GitHub stars in five months. It syncs the SKILL.md files
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai