
An official extension for Anthropic’s Claude mannequin was backdoored and threatened as many as 5,000 builders’ machines, a number of sources reported Friday. Anthropic thus far has but to warn clients of the incident.
The compromised bundle was a VS Code extension accessible within the market of Cursor, a maker of AI-assisted developer instruments, the corporate mentioned Friday. The malicious model, 2.1.313, subsequently affected “at most 5,000 customers,” presumably after they both downloaded it or their techniques routinely fetched and put in it. Anthropic pushed model 2.1.314 a number of hours after the earlier model turned accessible on Cursor, in line with impartial researcher Kevin Beaumont. Both Beaumont and the CTO and co-founder of safety agency Phorio
n mentioned that model 2.1.313 was distributed by Anthropic’s official account on Cursor. So far, neither Cursor nor Anthropic has defined how the attackers gained management of the account.
In search of cloud credentials
“It was a supply-chain compromise the place the attacker gained entry to the official distribution channel,” Phorion CTO and co-founder Luke Roberts instructed Ars. “The malicious model 2.1.313 was posted to the official Anthropic account, successfully superseding 2.1.312. Extensions auto-update in Cursor by default, however customers can configure this in settings.” Phorion has extra particulars right here.
Anthropic has but to problem public discover of the incident to clients.
Cursor mentioned that the malicious extension makes an attempt to put in a persistent SSH backdoor utilizing an added attacker-controlled SSH key, enabling the native SSH service, and opening a reverse SSH tunnel. The code was written to work on Windows, macOS, and Linux. Because Cursor extensions run with the permissions of the instrument’s Integrated Development Environment, the backdoored model possible would achieve entry to native recordsdata, supply code, and developer tooling as soon as it was put in.