The vulnerability in vm2 prior to version 3.11.8 arises because the allowAsync: false option does not fully prevent asynchronous execution within the sandbox. While the local Promise.prototype.then method is replaced with a handler that throws an error, the Promise static methods (Promise.resolve, Promise.all, Promise.race, Promise.any, and Promise.allSettled) still process attacker-supplied thenables. Native promise resolution invokes the sandboxed then method asynchronously in a microtask without passing through the patched then, allowing asynchronous code execution to continue after the sandboxed VM's run method has returned. This results in sandboxed scripts being able to schedule work beyond the configured timeout, potentially violating execution constraints.

Sandboxed code can bypass the intended asynchronous execution restriction and continue running after the host environment believes execution has completed. This can lead to unexpected code execution beyond timeouts, potentially causing resource exhaustion or other unintended side effects in the host environment. The vulnerability does not directly allow privilege escalation or remote code execution outside the sandbox but undermines the sandbox's execution control guarantees.

A fix is available in vm2 version 3.11.8 that fully enforces the allowAsync: false option. Users should upgrade to version 3.11.8 or later to remediate this issue. Until patched, users should be aware that asynchronous execution restrictions may not be fully effective and consider additional controls or monitoring as appropriate.