New security flaws allow data extraction from AMD virtual machines

Executive Briefing
“Researchers have uncovered two new attack methods named Mior and Skyfall, allowing...”
Key Takeaways
- 1.The Mior method developed by researchers from ETH Zurich allows intervention during virtual machine startup, enabling data reading and modification of memory as well as forging software attestation reports.
- 2.The Skyfall method conducted jointly by experts from ETH Zurich and Google exploits transient caching behavior to recover register values and disk encryption keys.
- 3.AMD has acknowledged these vulnerabilities and published a schedule to provide firmware and microcode patches for multiple EPYC processor lines in the coming months.
Researchers have uncovered two new attack methods named Mior and Skyfall, allowing malicious host servers to bypass memory encryption on AMD EPYC virtual machines protected by SEV-SNP. These vulnerabilities exploit the boot sequence or instruction timing to read sensitive data inside the virtual machines.
Key Developments and Technical Context
The Mior method developed by researchers from ETH Zurich allows intervention during virtual machine startup, enabling data reading and modification of memory as well as forging software attestation reports.
The Skyfall method conducted jointly by experts from ETH Zurich and Google exploits transient caching behavior to recover register values and disk encryption keys.
AMD has acknowledged these vulnerabilities and published a schedule to provide firmware and microcode patches for multiple EPYC processor lines in the coming months.
Strategic Analysis and Industry Implications
These attacks highlight the complexity of securing cloud environments for confidential computing technologies, especially when host operators control foundational firmware layers.
The reliance on cloud providers and hardware manufacturers to apply BIOS patches creates a significant challenge for customers renting confidential virtual machines.
Outlook and Key Factors to Watch
There is no confirmation from security advisories as to whether these vulnerabilities have been exploited in real-world attacks.
The specific effectiveness and methods for verifying trusted computing base values for virtual machine attestation after updates have not been detailed.







