A tampered version of Anthropic's Claude Code extension was posted to the extension marketplace of Cursor, an AI coding tool, The Verge and Ars Technica report. The bad version was available for about 11 hours on Thursday, Cursor said.

Cursor says around 5,000 people may have installed it, The Verge reports. Ars Technica quotes the company as saying "at most 5,000 users" were affected. The bad release was version 2.1.313, Ars reports. Extensions in Cursor update on their own by default, a security expert quoted by Ars said.

The code collected details from users' computers, including user names, home folder paths and public IP addresses, The Verge reports. Cursor said the code also tried to install a hidden back door that would let attackers reach the machine remotely, Ars Technica says. It was built to work on Windows, Mac and Linux computers.

Bugs in the code kept the back door from working in some cases on Windows and Mac machines, Ars Technica reports. Even so, the security firm Phorion called the malware "a significant threat."

Cursor removed the bad version and verified that "no other extensions were tampered with," The Verge reports. The company also cut off the server the back door connected to, Ars Technica says. It is still not known how attackers got into the verified account used to post the extension. Neither company has explained how, Ars reports.

Anyone who installed or updated the extension from Cursor on Thursday should consider their computer affected, Ars Technica says.

What we know: A tampered Claude Code extension was live on Cursor's marketplace for hours Thursday.
What we don't: How the attackers got access, or how many machines were actually compromised.