Critical Linux KVM flaw lets cloud VMs escape and compromise host systems

Researchers have disclosed a serious vulnerability in KVM, the virtualization layer built into many Linux systems, that permits guest virtual machines to escape isolation and potentially seize control of the underlying host hardware. The flaw, designated CVE-2026-53359 and nicknamed Januscape, affects KVM implementations running on both AMD and Intel processors and remained hidden in the kernel codebase for over a decade and a half.

How the vulnerability works#

The weakness stems from a use-after-free bug in the shadow MMU emulation layer—the component responsible for translating memory addresses between the hypervisor and guest systems. By executing malicious code entirely within the guest VM's own context, an attacker can corrupt the host kernel's shadow page tables, data structures essential for memory address translation.

Hyunwoo Kim, the security researcher who uncovered the flaw, released proof-of-concept code demonstrating how guest-side actions alone can trigger a kernel crash on the host system. He documented his findings alongside the discovery, noting that while a full remote code execution exploit exists, it will not be made public in the near term.

Threat to cloud infrastructure#

The implications for cloud platforms are severe. An attacker who rents a single virtual machine instance on a shared server could leverage this flaw to either mount denial-of-service attacks that knock other tenants offline, or execute code with administrative privileges to commandeer the entire physical machine and all guest systems running on it. This makes the vulnerability particularly dangerous in multi-tenant environments where isolation is essential.

Recognition and disclosure#

Google has awarded a $250,000 bounty for the discovery, reflecting the severity of the risk. The flaw surfaces alongside a second high-severity vulnerability in Linux disclosed this same week, indicating an active period of kernel security research.