AMD SEV‑SNP Vulnerabilities Expose Cloud Host Leaks
Two newly disclosed exploits let cloud hosts read or modify data inside AMD SEV‑SNP virtual machines, undermining the isolation promised by the technology.
On Friday, November 13 2026, The Hacker News reported that two new attacks—named Mior and Skyfall—have been made public this week. They allow a malicious host to read data stored inside AMD’s Secure Encrypted Virtualization – Secure Nested Paging (SEV‑SNP)-protected virtual machines (VMs). These VMs are designed to keep their memory encrypted from the server operator, so the leaks represent a serious breach of privacy for users who rely on the promise that only the VM itself can see its secrets.1
Mior works against an AMD SEV‑SNP VM by exploiting timing side channels. By repeatedly reading a protected region until the desired data appears, it extracts information without triggering obvious errors. In addition to reading, Mior can also alter a VM’s memory and forge the signed report that confirms which software is running in the machine. This capability effectively lets an attacker impersonate legitimate code inside the VM, potentially allowing further lateral movement within the guest operating system. The attack requires control over the hypervisor and, for Mior specifically, control of the server BIOS as well.1
Skyfall takes a different route. It abuses vulnerabilities found in Intel VT‑d IOMMU drivers, enabling an attacker controlling the host firmware to bypass security mechanisms entirely. Because the flaw lies in low-level driver code, Skyfall does not trigger detection mechanisms that would normally flag abnormal activity; instead it silently sidesteps protections meant to isolate the VM from the host. Like Mior, Skyfall needs hypervisor control but does not require BIOS access. Researchers note that Microsoft has been battling similar issues for years in Hyper‑V environments, where analogous privilege escalation paths have emerged.1
Both attacks exploit weaknesses around attestation—the process by which a VM proves its integrity to external parties. AMD’s SEV‑SNP aims to provide strong isolation by encrypting each VM’s memory and preventing the hypervisor from inspecting or tampering with it. However, the new techniques demonstrate that if an attacker can influence the hypervisor (or firmware), they can circumvent even the most robust encryption schemes. This undermines the core promise of SEV‑SNP: that only the guest operating system and its authorized users can read sensitive data.
Many enterprises deploy VMs on shared hardware under the assumption that SEV‑SNP protects them from co-tenant snooping. With Mior and Skyfall, those assumptions no longer hold—any malicious host could exfiltrate secrets stored inside the protected VM. Even though the attacks do not immediately grant full OS compromise, their ability to read and modify encrypted memory is enough to undermine confidentiality guarantees.
Cloud operators must now consider patching or reconfiguring their infrastructure. While the exploits target specific kernel modules and driver bugs, mitigating requires updating firmware, disabling vulnerable IOMMU paths, or moving workloads away from affected hardware until vendor patches become available. The research community has called for a review of SEV‑SNP’s attestation protocol, noting that future designs should include stronger checks against timing leakage and driver manipulation.1
In short, the Mior and Skyfall disclosures reveal that AMD’s latest virtualization security technology still leaves room for determined attackers with hypervisor control to access supposedly sealed VM memories. Users who depend on SEV‑SNP for privacy must stay alert to these new threats and await official fixes before trusting it for sensitive workloads.
Made by Ren, the ArRENCE AIChat with Ren, plan your day, generate images, and more.
Meet Ren →