Firecracker VM Escape: PwnEscape Part 2

258 points · 110 comments on HN · read original →

Points and comments are a snapshot, not live.

Researchers escape Firecracker microVM and execute code as root on the host.

The exploit chain uses host memory access to find Firecracker's process in memory, locate a vCPU thread's saved registers, and modify them to redirect execution from KVM_RUN. By altering the pt_regs frame, the guest makes Firecracker call write(2) with an attacker-chosen string. The target was Firecracker v1.16.1 on x86-64 with host kernel 7.0.0-1011-aws, run as root without a jailer. The guest's serial write triggers the return, and the payload executes with Firecracker's privileges. AWS published a security bulletin (2026-137) stating affected services are not impacted.

What commenters are saying

Some commenters dismiss the article as LLM-generated slop, but others defend its substance as the first general-purpose Firecracker escape without nested virtualization. Several point out that a properly jailed Firecracker deployment would reduce the blast radius. Discussion notes the root cause is a kernel network code flaw (checksum offload) fixed upstream. A user running ClickBench Playground on Firecracker VMs offers to PoC the escape on their service. Links to AWS's security bulletin and the earlier PwnEscape Part 1 disclosure are shared.