A recent security breach has raised significant concerns as an official extension for Anthropic’s Claude AI model was compromised, potentially impacting as many as 5,000 developers’ machines. Reports surfaced on Friday detailing the incident, although Anthropic has yet to inform its customers about the threat.
The malicious extension, identified as version 2.1.313, was available on the marketplace of Cursor, a developer of AI-assisted tools. The breach may have affected users who either downloaded the extension or had their systems automatically update. Following the incident, Anthropic quickly released a revised version (2.1.314), just hours after the malicious version was made public. Independent researcher Kevin Beaumont noted that the compromised version was disseminated via Anthropic’s official account on Cursor, yet neither Cursor nor Anthropic have provided clarity on how the incident occurred.
Supply Chain Vulnerability
According to Luke Roberts, CTO and co-founder of security firm Phorian, this incident exemplifies a supply-chain compromise where the attacker accessed an official distribution channel. The malicious extension was uploaded under Anthropic’s account, replacing version 2.1.312. By default, Cursor extensions are set to auto-update, although users can adjust this in their settings. For further insights, Phorian has provided additional information on their blog.
The compromised extension is designed to create a persistent SSH backdoor, utilizing an attacker-controlled SSH key. This allows unauthorized access to local SSH services and establishes a reverse SSH tunnel. The backdoored code is compatible with Windows, macOS, and Linux platforms. Due to the permissions held by Cursor extensions within the respective Integrated Development Environment, the malicious software could potentially access sensitive local files, source code, and developer tools.
So far, there has been no public acknowledgment of this security incident from Anthropic.
Editor’s Take
This incident underscores the critical vulnerabilities within software supply chains, especially for tools used by developers. With growing reliance on AI-assisted solutions, the implications could affect not just individual users but also organizations that leverage these technologies. Enhanced security measures and transparency are urgently needed to protect against such breaches in the future.
Source: arstechnica.com
