Anthropic Account Breach Exposes Escalating Risks in AI Developer Tooling

A malicious version of Anthropic’s official VS Code extension, distributed via Cursor, compromised up to 5,000 developer machines. The breach installed persistent SSH backdoors, yet both companies have remained silent on how the account was accessed.

AI & Culture Desk
Updated 3d ago 2 min read
Anthropic Account Breach Exposes Escalating Risks in AI Developer Tooling

A supply-chain compromise allowed attackers to distribute a backdoored version of Anthropic’s official VS Code extension through the Cursor platform, potentially exposing as many as 5,000 developers to persistent unauthorized access. According to Ars Technica, the malicious update, version 2.1.313, was active on Thursday, between 5:27 am ET and 4:35 pm. The incident highlights a significant security gap in AI-assisted coding tools, as the extension was distributed directly through Anthropic’s verified account, bypassing the standard trust mechanisms that developers rely on when installing software.

The malicious code was designed to operate across Windows, macOS, and Linux environments. Once installed, the extension attempted to establish a persistent SSH backdoor by adding an attacker-controlled key and opening a reverse SSH tunnel. Security firm Phorion, which analyzed the incident, noted that the extension’s capabilities included the theft of variable names, allowing attackers to inventory cloud credentials and API tokens stored on the host machine. While some bugs in the malicious code prevented the backdoor from functioning on certain Windows and macOS systems, Phorion emphasized that the malware remained a significant threat by beaconing host reconnaissance back to the operator.

Cursor’s response involved coordinating with the cloud provider hosting the attacker’s server to sever the reverse shell connection, effectively neutralizing the immediate threat to infected devices. The company subsequently removed the malicious version from its marketplace and temporarily shut down the VSCode Marketplace to rotate security keys and conduct a comprehensive audit of access controls. Despite these technical remediations, the incident has raised questions regarding transparency. Ars Technica reports that Anthropic has not issued a public notice to its customers, and neither Anthropic nor Cursor has provided an explanation regarding how the attackers gained control of the official distribution channel.

The breach underscores the elevated risk profile of AI-centric developer tools. Because these extensions function with the permissions of the Integrated Development Environment (IDE), they possess deep access to local file systems, private source code, and sensitive developer tooling. By compromising a verified publisher account, attackers can bypass traditional perimeter defenses and deliver malicious payloads directly to trusted environments. The fact that the malicious version 2.1.313 superseded the legitimate version 2.1.312—and that extensions in Cursor auto-update by default—demonstrates how quickly a supply-chain attack can propagate through a developer ecosystem.

For engineering organizations, this incident serves as a stark reminder of the vulnerabilities inherent in modern AI tooling stacks. While Cursor has restored marketplace operations, the lack of disclosure from Anthropic regarding the breach vector leaves a critical gap in understanding how such high-value accounts are secured. Until labs and platform maintainers provide greater clarity on their security protocols and how they verify publisher integrity, developers remain exposed to similar risks. Users who installed or updated the extension during the affected window are advised to consider themselves potentially compromised and should review indicators of compromise provided by security researchers.

Sources

  1. 01 Anthropic VS code extension on Cursor backdoored for nearly 12 hours — Ars Technica