vm2 Sandbox Bypass via Internal HTTP Builtins

vm2 is a popular Node.js sandbox that restricts access to dangerous network modules like http, https, and net when configured with wildcard builtin exclusions. However, this vulnerability reveals a critical gap in that protection: Node.js exposes internal underscore-prefixed modules (_http_client, _http_server, _http_agent, and others) that provide equivalent functionality but weren't blocked by the public module filtering. This allows sandboxed code to bypass network restrictions entirely, enabling attackers to establish outbound connections and exfiltrate data despite the sandbox's intended isolation. Any application using vm2 to run untrusted scripts—including multi-tenant platforms, code execution services, and security research tools—faces direct risk of sandbox escape and data theft.

Casky's approach to detecting this attack pattern focuses on behavioral analysis rather than signature matching. Practitioners using Casky would identify anomalous module loading patterns: specifically, attempts to require underscore-prefixed or internal Node.js modules that bypass the intended builtin allowlist. While MITRE ATT&CK mappings aren't provided for this CVE, the underlying attack chain maps to Defense Evasion (T1548 - Abuse Elevation Control Mechanism) and Exfiltration (T1041 - Exfiltration Over C2 Channel). Claude's extended reasoning capabilities would help security teams understand the privilege escalation within sandbox contexts and recognize how legitimate-looking require() calls to internal modules represent a fundamental sandbox escape. Practitioners would see findings highlighting suspicious module resolution patterns and comparing actual loaded modules against configured allowlists—surfacing the critical mismatch that defines this vulnerability.