Two novel attacks, dubbed “Mior” and “Skyfall,” have emerged that allow malicious cloud hosts to bypass the protections offered by AMD’s Secure Encrypted Virtualization with Secure Nested Paging (SEV-SNP). These vulnerability disclosures signal serious risks to VM confidentiality and integrity—even when using hardware security features meant to isolate guest VMs from the host or hypervisor.
What are Mior and Skyfall?
Mior enables an attacker who controls both the hypervisor and the BIOS (UEFI) of an AMD EPYC server to tamper with the Reverse Map Table (RMP) during the system’s startup process. The RMP, initialized by the Platform Security Processor (PSP), maps memory pages and defines which software components can access them. In lab tests, researchers rerouted the PSP’s writes to a controlled device (a graphics card), leaving parts of the real RMP uninitialized. This allowed post-boot manipulation of memory protections, enabling the attacker to enable debugging, read and write VM memory, and forge attestation reports. All of this worked reliably across multiple tests with ciphertext hiding enabled.
Skyfall takes a different route: it leverages the Transient Scheduler Attacks (TSA) class of timing side-channel flaws to extract data from VMs. Specifically, when a VM fails to initialize, the processor can leave behind register and state data in cache without proper cleanup. The hypervisor, using TSA, can then extract parts of saved state—including secret disk encryption keys and register values—by forcing failure and observing timing. Skyfall also leaks data from fixed cache lines at certain memory offsets. Unlike Mior, Skyfall does not require BIOS tampering, only hostile control of the hypervisor.
Affected Systems and Fixes
The malware team tested Mior using a Zen 5 EPYC 9135 processor and viewed Zen 3 and Zen 4 chips as likely vulnerable though not directly tested due to lack of BIOS modification capabilities. Skyfall was validated on Zen 3 (EPYC 7313) and Zen 4 (EPYC 9124, EPYC 9334) systems; tests on Zen 5 chips did not reproduce the VM memory leak. In addition to VMs, Skyfall also exposed decrypted microcode updates and System Management Mode (SMM) state.
AMD has assigned CVEs to both attacks: Mior is CVE-2026-0462 with a 5.9 severity (CVSS 4.0), while Skyfall’s VM leak is CVE-2026-40671 with a score of 4.9. Vendor advisories list fixes via platform firmware (BIOS/UEFI) and SEV firmware updates. Skyfall’s patch microcode is dated August 12 for affected lines, with BIOS rollouts set for November; Mior’s full platform firmware patches are expected in December. Users must update both firmware and microcode, but availability depends heavily on cloud providers or server OEMs to deploy correctly.
Why This Matters
SEV-SNP is designed to protect VMs even from compromised hypervisors via hardware memory encryption, access control, and attestation of software state. These new attacks show that under certain threat models—control of hypervisor + BIOS in Mior, or hypervisor alone in Skyfall—these guarantees can be undermined. Attacks like Mior even let an adversary fake which software the VM claims to run.
What Should Users & Providers Do?
First priority: deploy vendor firmware updates and BIOS/UEFI patches as soon as they are available. For Mior, that may include significant BIOS work; for Skyfall, microcode updates must be applied, as well as host firmware. Second: for cloud operators, ensure Trusted Computing Base (TCB) values shown in attestation match patched firmware versions. For VM owners using confidential computing, check with cloud providers that hardware firmware is up to date.
Neither advisory claims these vulnerabilities are known to have been exploited yet. Proof-of-concept code for Mior has been published, though it requires BIOS configurations that leave memory-routing settings unlocked.
These attacks underscore that even advanced hardware protections like SEV-SNP rely critically on secure initialization and firmware integrity. The threat model promised by confidential computing vendors must assume supply chain, firmware, and hypervisor integrity. Going forward, transparency about firmware versions, attestation guarantees, and testing of patch efficacy will be key in defending cloud confidentiality.
