Execute arbitrary code inside a sandbox via protection mechanism failure in V8
No working AI exploit yet