Threat level Guarded
Submit a tip

Major threats, straight to your inbox

Only the threats that matter: actively exploited flaws, big breaches and attacks that need action. No spam, unsubscribe any time.

Mior and Skyfall Attacks Break AMD SEV-SNP Confidential VM Protection

ETH Zurich's Mior (CVE-2026-0462) lets a rogue host leave AMD SEV-SNP's Reverse Map Table unprotected, read and write confidential VM memory and forge attestation, while Skyfall (CVE-2026-40671) leaks VM register state and a disk key. AMD fixes ship through OEM BIOS updates in November and December.

Mior and Skyfall Attacks Break AMD SEV-SNP Confidential VM Protection

Threat card

Threat type
Cloud Security
Disclosed
Nov 13, 2026
Affected
AMD EPYC 7003, 8004, 8005, 9004 and 9005 server processors running SEV-SNP confidential VMs; Skyfall also affects Instinct MI300A and embedded EPYC lines

The short version

  • Mior lets a malicious host with BIOS control redirect PSP writes to a PCIe device, leave the RMP partly uninitialised and then read, write and forge attestation for SEV-SNP VMs.
  • Mior worked in all 1,000 attempts on an EPYC 9135, even with ciphertext hiding enabled.
  • Skyfall abuses Transient Scheduler Attacks on deliberately failed VM starts to leak register state, including a disk-encryption key in a test.
  • AMD fixes arrive through OEM BIOS and microcode updates, so confidential VM tenants depend on their cloud provider to patch.

Two research attacks disclosed this week, Mior and Skyfall, show that a malicious cloud host can break the guarantees of AMD SEV-SNP, the hardware feature behind confidential virtual machines. Mior (CVE-2026-0462) lets a rogue host read and write confidential VM memory and forge attestation reports, succeeding in 1,000 out of 1,000 attempts. Skyfall (CVE-2026-40671) leaks VM register state, including a disk-encryption key in testing. AMD fixes reach customers through OEM BIOS updates in November and December 2026, so tenants of confidential VMs need to ask their providers when hosts are patched.

What happened

Mior and Skyfall are hardware-level attacks against AMD SEV-SNP that assume a malicious host or hypervisor, the exact threat confidential computing is meant to stop. Mior comes from ETH Zurich and Skyfall from ETH Zurich and Google. AMD published bulletins SB-3059 for Mior and SB-3061 for Skyfall, and neither advisory reports exploitation in the wild.

The attacks matter because SEV-SNP underpins confidential VM offerings that companies use for sensitive data and confidential AI workloads. The Mior researchers note that ordinary VMs already trust their host and firmware, so Mior gives an attacker nothing new against them. The damage is to the promise that a cloud operator cannot see inside a protected VM.

Timeline

Date Event
8 July 2025 Transient Scheduler Attacks (TSA), the basis for Skyfall, first disclosed
12 August 2026 Date of AMD SEV firmware for Mior and Skyfall microcode for affected EPYC lines
21 September 2026 Skyfall patch dated for Instinct MI300A
Week of 13 November 2026 Mior and Skyfall publicly disclosed with AMD bulletins
November and December 2026 Platform firmware packages target dates for EPYC server lines
January to April 2027 Fixes due for embedded EPYC processors

How the attack works

Mior: an unprotected Reverse Map Table

SEV-SNP relies on the Reverse Map Table (RMP) to stop the hypervisor from tampering with VM memory. According to The Hacker News analysis of both papers, the Mior team redirected the Platform Security Processor’s writes for the RMP to a graphics card they controlled. SEV-SNP startup completed without errors, but part of the RMP, including the entry protecting the table itself, stayed uninitialised.

With that gap, the hypervisor could rewrite RMP protections. The researchers enabled debugging on a confidential VM that had it disabled, then used PSP debug commands to read and write its memory. In a second test they copied a harmless VM’s launch measurement into a malicious VM, so attestation reports described the harmless image. The Mior paper reports success in all 1,000 attempts, and the attacks still worked with ciphertext hiding enabled.

Mior requires control of the BIOS or UEFI and a PCIe device. The researchers point out that “AMD explicitly distrusts the UEFI, thus an attacker can modify the UEFI,” so this is inside SEV-SNP’s own threat model.

Skyfall: register state from failed VM starts

Skyfall builds on Transient Scheduler Attacks. When a hypervisor deliberately makes a VM fail to start, the processor has already cached part of the VM’s saved state, and that data is left without the usual cleanup. TSA side channels can then read it.

The researchers leaked saved VM register values, three fixed cache lines from a targeted page, decrypted microcode updates and register values in System Management Mode memory. They recovered a disk-encryption key from a test VM by pausing its kernel when the key sat in registers, and say SEV-Step could be used to reach that point in a real attack. The IEEE S&P programme committee noted that the paper overlooks what a realistic attack would require, such as an otherwise idle core and a VM that does not crash or time out.

MITRE ATT&CK mapping

Tactic Technique (ID) How it was used here
Persistence Pre-OS Boot: System Firmware (T1542.001) Mior requires attacker-modified BIOS or UEFI on the host
Defense Evasion Subvert Trust Controls (T1553) Forged SEV-SNP attestation reports describe a harmless VM
Credential Access Unsecured Credentials (T1552) Skyfall leaks a disk-encryption key from VM registers
Collection Data from Local System (T1005) Mior reads confidential VM memory through PSP debug commands

Who is behind it

These are academic disclosures, not attacks by a threat actor. Mior is from ETH Zurich and is separate from the earlier Fabricked and RMPocalypse attacks by some of the same researchers; it also worked on firmware predating the Fabricked fix. Skyfall is from ETH Zurich and Google, and its paper was accepted to the IEEE Symposium on Security and Privacy. Mior proof-of-concept code is public on GitHub.

Affected products and versions

Product Affected Fixed CVE CVSS
EPYC 7003 (Milan) SEV-SNP MilanPI-SP3_1.0.0.L; SEV firmware 1.3A.03 CVE-2026-0462, CVE-2026-40671 5.9 / 4.9 (CVSS 4.0)
EPYC 8004 and 9004 (Genoa) SEV-SNP GenoaPI_1.0.0.K (Mior), Genoa_100J Patch 3 (Skyfall); SEV firmware 1.3A.4 CVE-2026-0462, CVE-2026-40671 5.9 / 4.9 (CVSS 4.0)
EPYC 8005 and 9005 (Turin) SEV-SNP TurinPI_1.0.0.C; SEV firmware 1.3A.0A CVE-2026-0462, CVE-2026-40671 5.9 / 4.9 (CVSS 4.0)
EDK II based firmware (SMM leak) Missing TSA mitigations EDK II patch per AMD SB-7072 CVE-2026-18448 5.6

Mior was tested only on an EPYC 9135 (Zen 5); Zen 3 and Zen 4 are considered potentially affected. Skyfall was shown on EPYC 7313, 9124 and 9334 but not on Zen 5. AMD lists only SEV-SNP as affected, not SEV or SEV-ES, and no workarounds are listed.

Indicators of compromise

These are research disclosures, so no IOCs exist. Exploitation would happen on the host, out of the tenant’s view. Signals worth tracking are attestation reports showing a TCB version below AMD’s fixed values, unexpected debug policy on a confidential VM, and repeated failed VM launches on a host, which Skyfall relies on.

Detection and hunting

Tenants should log and review every SEV-SNP attestation report their workloads produce and compare reported TCB values against the minimums in AMD’s bulletins as providers adopt them. Any VM whose policy shows debugging enabled when it was launched without it should be treated as compromised.

Operators of self-managed EPYC hosts should monitor BIOS and BMC integrity, unexpected PCIe device changes and spikes in VM launch failures. The attacks fit a wider pattern of cloud trust failures Threatspedia has tracked, including the Qubole AWS trusted-role breach.

Mitigation and response

  1. Install AMD platform firmware, SEV firmware and microcode from your server maker; Supermicro has published a BIOS fix list and says BMC updates are also needed.
  2. Cloud tenants should ask providers for a written date when hosts running confidential VMs receive the SB-3059 and SB-3061 fixes.
  3. Enforce minimum TCB versions in attestation policies once fixed values are known for your platform.
  4. Apply the EDK II SMM fix from AMD SB-7072 where relevant.
  5. Until hosts are patched, avoid relying on SEV-SNP alone for the most sensitive keys and add application-level encryption.

Analyst take

Confidential computing sells one idea: you do not have to trust the cloud operator. Mior shows that a host with firmware control can still take a confidential VM apart and lie about it in attestation, which is the control customers use to check that promise. The low CVSS scores reflect the strong preconditions, not the size of the trust failure.

The patch path is the real problem. Fixes travel from AMD to OEMs to cloud providers, and tenants cannot patch the host themselves. Teams running confidential AI or regulated workloads should make attestation TCB checks mandatory and treat provider patch timelines as a contract issue, not a footnote.

What to do now

0 of 5 done

Questions readers ask

What is the Mior attack?

Mior (CVE-2026-0462) lets a malicious host with BIOS control leave AMD SEV-SNP's Reverse Map Table unprotected, then read and write confidential VM memory and forge attestation reports.

What does Skyfall leak?

Skyfall (CVE-2026-40671) leaks saved VM register values through Transient Scheduler Attacks on failed VM starts; researchers recovered a disk-encryption key from a test VM.

How are Mior and Skyfall fixed?

AMD ships firmware and microcode fixes that reach customers through server maker BIOS updates; cloud tenants must rely on their provider to install them.

Sources

  1. AMD: SB-3059 Mior security bulletinamd.com
  2. AMD: SB-3061 Skyfall security bulletinamd.com
  3. ETH Zurich: Mior paper (CCS 2026)xca-attacks.github.io
  4. Skyfall paperskyfall-attack.github.io
  5. The Hacker News: New Mior and Skyfall attacks let rogue hosts break AMD SEV-SNPthehackernews.com
Share this story
Get the next major threat first

One short email when something serious breaks.

Threatspedia Desk

Threatspedia covers major cyber threats only, verified against primary sources. Read our editorial policy, or report a correction.

Threat alerts

Be first to know when a major threat breaks.

One short email when something serious happens, plus a weekly roundup. Free, no noise.