Malicious Cursor/Claude Code Extension Triggers Recon, SSH Tunnel Attempt

En bref

• Unknown attacker used a malicious extension in a Cursor/Claude Code development setup.
• Activity included host reconnaissance and an attempted reverse SSH tunnel.
• No CVE was cited; the number of affected systems is not yet specified.
• High severity due to the tunnel attempt and access to developer hosts.

An unidentified threat actor was involved in an incident where a harmful add-on was introduced into a workflow that combined Cursor's integrated development environment with Anthropic's Claude Code extension. The activity did not rely on a published software flaw, and no CVE identifier has been associated with the event. Instead, the reported behavior centers on the extension itself and the access it obtained on the developer's machine.

After delivery, the extension carried out discovery actions against the host, according to the available facts. It also tried to establish a reverse SSH tunnel, a technique that can create an outbound path from the compromised system to an external endpoint. That combination is significant because reconnaissance maps the environment while a tunnel can provide persistent remote access if it succeeds.

The exact number of affected users or systems has not been disclosed. The exposure is most relevant to developers and IT teams using Cursor and the Claude Code extension in K-12 or government environments, where developer workstations often hold credentials, source code, and access to internal services. Even a single compromised endpoint can become a stepping stone into broader systems.

Because no CVE is involved, traditional vulnerability scanning may not flag this event. Detection depends on extension inventory, endpoint telemetry, and network monitoring for unusual outbound SSH behavior. Organizations should treat the report as a supply-chain and tooling-risk issue rather than a patchable software defect.

What to watch: unexpected extension installations or updates, unusual child processes spawned by the IDE, outbound connections on SSH ports to unfamiliar hosts, and signs of host enumeration such as user, network, or service discovery commands. Administrators should preserve logs and validate whether any tunnel attempt succeeded.

À faire maintenant

1. Isolate any developer endpoint where the malicious extension was observed; preserve volatile data and logs before cleanup.
2. Inventory Cursor and Claude Code extensions across developer fleets; remove unapproved or unknown packages and verify publisher and source.
3. Review endpoint and network logs for host discovery commands and outbound SSH connections, especially reverse-tunnel patterns, from IDE-related processes.
4. Block or alert on unauthorized outbound SSH and restrict egress from developer workstations to approved destinations.
5. Rotate credentials, API tokens, SSH keys, and session tokens that were accessible from affected hosts.
6. Enforce an allowlist for IDE extensions and require approval before installation; disable auto-install where possible.
7. Report the incident to security operations and vendor contacts; monitor for follow-on access using stolen credentials.

Source originale

SecurityWeek