العربية
🚨 BreakingFBI Warns Artex AI Tool Automates Hacks at Machine Speed

AMD EPYC Flaws: Mior and Skyfall Break SEV-SNP Isolation

Security World Teamاقرأ بالعربية

Two new attacks, Mior and Skyfall, let a malicious host read data from VMs protected by AMD SEV-SNP. The same week, Google Cloud told AMD Turin customers to patch three side-channel flaws.

Illustration for Vulnerabilities & CVEs
Illustration by Security World

⚡ Why it matters

SEV-SNP underpins confidential computing in the cloud, where customers are promised protection even from the server operator. Both attacks need deep privileges, but they weaken that promise.

AMD SEV-SNP encrypts a virtual machine's memory and isolates it even from the hypervisor. This week, two attacks showed lab bypasses of parts of that protection on AMD EPYC processors.

The two attacks

  • Mior (ETH Zurich) tampers with the boot process so part of the RMP memory-protection table is never initialized. The team read and modified a confidential VM's memory and forged its attestation report. It requires control of the hypervisor, the BIOS and a PCIe device such as a graphics card. AMD tracks it as CVE-2026-0462 (CVSS 5.9).
  • Skyfall (ETH Zurich and Google) builds on Transient Scheduler Attacks (TSA) and leaks a few fixed cache lines from a VM page. AMD tracks the VM leak as CVE-2026-40671 (CVSS 4.9).

AMD lists firmware and microcode dated August 12 and plans to ship platform firmware to server makers in November for Skyfall and December for Mior.

Google Cloud advisory

Google said VMs on AMD Turin (5th-gen EPYC) in the C4D, N4D, H4D, Z4D and G4 series are affected by three TSA flaws, including the new CVE-2026-0004. Google has mitigated cross-VM attacks, but customers must install their OS fix and then stop and start the VM, since a guest reboot is not enough. No exploitation has been reported.

What to do

  1. Get the BIOS update from your server vendor as soon as it ships.
  2. On Google Cloud, patch the guest OS, then stop and start the VM.

🔗 Sources

This article is original reporting by the Security World editorial team based on the sources listed above; direct quotes are kept short and marked with quotation marks. Corrections and feedback: info [at] securitywd.net.

Related stories