Developers installing an official Claude extension through Cursor’s marketplace received software designed to give attackers lasting remote access to their machines, Ars Technica reported on Oct. 30.[1] The compromised software was Anthropic’s Claude Code extension for VS Code, distributed through Cursor, a maker of AI-assisted developer tools.[2][1]

The danger extended to developers who made no deliberate installation: Cursor updates extensions automatically by default, security firm Phorian’s CTO and co-founder Luke Roberts told Ars Technica.[1] Ars Technica noted that extensions inherit the editor’s permissions, potentially exposing local files, source code and developer tools.[1]

According to Microsoft Security Blog’s analysis, the injected code runs when the extension loads and attempts to authorize attacker access, establish a remote connection and configure scripts to restore that connection.[3] The extension continues functioning normally while concealing malicious activity.[3] The code sends selected configuration-variable names, not their values; Microsoft said this does not demonstrate theft of the secrets stored there.[3]

Microsoft Security Blog said the malicious packages never reached Microsoft Visual Studio Marketplace or npm, and reported that Anthropic told Microsoft its systems were not compromised.[3]

The sources differ on affected versions: Ars Technica identified 2.1.313, while Microsoft Security Blog also listed 2.1.312.[1][3]

Cursor removed the malicious extension, blocked distribution and worked with the hosting provider to cut infected devices off from the attacker’s server, Ars Technica reported.[1] But Anthropic had issued no public customer warning, and neither company answered questions about how the compromise happened.[1]