‘Fragile security model’
These sandbox escape vulnerabilities demonstrate why sandboxing untrusted code inside a trusted process is a fragile security model, Adam Reynolds, senior security researcher at Sonatype, said in an email. “Once untrusted code runs inside a process with access to credentials and secrets, the underlying filesystem, the network, or with deployment privileges, a sandbox bypass can easily lead to a full system compromise,” he said.
Simply having vm2 installed somewhere in the dependency tree is not enough to make some of these vulnerabilities exploitable, he added. For example, an attacker generally needs the ability to execute crafted JavaScript (and in the case of CVE-2026-26956, crafted WebAssembly) inside a vm2 sandbox controlled by the vulnerable application. If the application never instantiates vm2, only uses it for trusted internal scripts, or does not allow attacker-controlled code execution at all, then there may be no realistic exploit path despite the presence of the dependency.
