Firecracker VM Escape: PwnEscape Part 2

PWN Research published the second part of its PwnEscape series, a guest-to-host escape from Firecracker, the microVM monitor widely used to isolate untrusted workloads. The technique starts from a guest that already has a primitive for reading and writing host memory. From there, the guest locates the VMM process through a nonce page and EPT address translation. It then rewrites the saved registers of a vCPU thread, so that when KVM_RUN returns, the thread executes attacker-chosen code on the host as root. That chain matters to agent builders because Firecracker is a common choice for running code that models write. The escape still needs an initial memory primitive, but the write-up shows that once one exists, the remaining path to host control is short and documented. Teams that rely on Firecracker isolation now have a concrete escape path to measure their patch levels and outer defense layers against.