SecurityNews
A Trusted Claude Code Update Contained a Backdoor. What Else Are We Missing?
Cursor says a tampered build of Anthropic's extension was live for about 11 hours on October 29, and nobody has explained how the official account was used.
What the reporting says
BleepingComputer reports that a tampered build of Anthropic's Claude Code extension, version 2.1.313, sat in Cursor's marketplace for roughly 11 hours on October 29. Cursor says the code was meant to plant a lasting SSH backdoor and that at most 5,000 users were exposed. Ars Technica reports that independent researcher Kevin Beaumont and the CTO of security firm Phorian both say the build was published under Anthropic's official account, and that Anthropic shipped 2.1.314 several hours afterward. Cursor says Anthropic did not write the malicious code. According to Ars Technica, neither company has said how the attackers got into that account, and Anthropic had not yet warned customers. Dark Reading adds that Cursor dealt with a second serious security problem the same day.
Read the original reporting
BleepingComputer
Malicious Claude Code extension on Cursor planted SSH backdoorsMy Read
Commentary: opinion, not reporting.
I'm a Claude Code user, so this one has my attention. Developers installed an update from a trusted publisher, and that update contained a backdoor. The users did everything right. Apparently, that's no longer enough.
Here's what bothers me. Was this someone, or something, planting access for a future attack? Could autonomous AI agents be involved? We've already seen malicious AI tools and compromised software supply chains. Are these isolated incidents, or are we just seeing the tip of a new iceberg?
Nobody has explained how the malicious update got published. Meanwhile, we're giving AI coding tools increasingly broad access to our machines, legacy codebases, and credentials. We're building and publishing complex applications containing code we may not fully understand, while corporations aggressively look to reduce their reliance on software engineers. What could possibly go wrong?
Discussion
Should AI coding tools that run with broad access to a developer's machine get a stricter update process than ordinary editor extensions?
Covered by 4 of 22 outlets we read, on Techmeme, 936 points on Hacker News.