AI · SEMICONDUCTOR · FACT CHECK
SKD WIRETrue
The Anthropic VS Code extension running on the Cursor development environment was compromised by a backdoor that persisted for nearly 12 hours before being addressed, according to a review of the attack chain documentation and six primary sources.
The attack targeted the extension environment rather than the models themselves. Documentation of the attack chain shows the backdoor was planted in the Anthropic VS Code extension as used on Cursor, giving the attacker a foothold in a developer toolchain widely used for AI-assisted coding.
The compromised state lasted close to 12 hours. Within that window, the backdoor was live in the extension, though the full scope of affected users and the specific exfiltration or execution mechanisms remain unclear from the available documentation.
Details of the remediation — including when Anthropic or Cursor first detected the issue and what steps followed removal — have not been fully disclosed in the materials reviewed. The incident underscores how quickly AI coding integrations can become attack surfaces.
The nearly 12-hour backdoor in the Anthropic VS Code extension on Cursor is True.