Cursor finds SSH backdoor in fake Claude Code extension
Cursor disclosed a security breach on October 29, 2026, after detecting a malicious version of the Anthropic.claude-code VS Code extension in its marketplace. The package, version 2.1.313, was not developed by Anthropic and attempted to install a persistent SSH backdoor on up to 5,000 users. It ran between 2:27 a.m. PT and 1:35 p.m. PT on October 29, 2026, before Cursor removed it and shut down…
Key points
- Attempted SSH backdoor via reverse tunnel, persistence scripts, and data exfiltration to attacker’s **134.209.82.225:8099**
- Cursor shut down marketplace, removed malicious package, and provided remediation scripts for affected users
The extension collected system details—hostname, OS, Git email, and IP address—and tried to add an attacker-controlled SSH key to users’ authorized keys files. While the reverse SSH tunnel failed on most systems due to missing admin privileges, it still installed persistence mechanisms (LaunchAgent on macOS, scheduled tasks on Windows) and sent data to an attacker-controlled endpoint at 134.209.82.225:8099. Cursor confirmed the attacker’s server was taken offline, but users are advised to check for lingering backdoors using provided scripts and rotate credentials if affected. The company emphasizes the incident only involved the tampered extension and not the official Anthropic distribution.
The story so far
4 episodes →- Cursor finds SSH backdoor in fake Claude Code extensionthis story
Cursor Discloses SSH Backdoor Attempt in Malicious Claude Code Extension
Unite.AI · 30 October 2026
Loading the full article…
This text was published by Unite.AI and written by Miles Okada, AI & Cybersecurity, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Society & Work
All →- OpenAI outlines four-layer system to detect AI misalignment in advanced models · 4 src
- OpenAI CEO Altman says he feels conflicted about AI solving math problems · 1 src
- CrowdStrike says China-linked hackers used DeepSeek to target 200+ groups · 2 src
- Researcher wins first Title VII AI exemption for humanist beliefs · 2 src
- MIT professor Eric So warns AI reliance erodes human cognitive skills · 1 src
Comments
via GitHub Discussions