Mior and Skyfall break trust in AMD SEV-SNP
Two attacks allow a malicious host to read secrets from VMs protected with AMD SEV-SNP. We analyze the real risk for SMBs and what measures to take.
4 min read

In this article
What has happened with AMD SEV-SNP
This week, two attacks have been made public, dubbed Mior and Skyfall, which allow a malicious cloud operator to read data inside virtual machines protected with AMD SEV-SNP. This technology is designed precisely for the opposite: to prevent the server provider from being able to look inside the customer's VM. The researchers have demonstrated that this barrier is not insurmountable under certain conditions.
The important nuance lies in the requirements. Both attacks require the attacker to control the server's hypervisor. Mior goes one step further and also requires control of the machine's BIOS. In addition, Mior does not only read memory: it can modify it and forge the signed report that confirms what software is running inside the VM. In other words, we are not talking about a flaw that can be exploited from the internet just like that, but rather a scenario in which the infrastructure provider itself, or someone who has compromised its virtualization layer, has the ability to act against the tenant.

Why this matters to an SMB
At first glance, an attack that requires control of the hypervisor and the BIOS sounds like a problem for large cloud providers, not for a company with ten or fifty employees. And in part that is true: the probability that a top-tier cloud provider is compromised in that way is low. But there are three readings that do affect the day-to-day operations of an SMB and its IT team.
The first is trust in the cloud model. Many SMBs have migrated sensitive workloads (billing, customer data, intellectual property) to third-party infrastructure relying on the fact that memory encryption and confidential environments protect them even from the provider. News like this reminds us that this guarantee is one more layer, not an absolute railing. It is not a reason to run away from the cloud, but it is a reason not to delegate all security to the provider's promise.
The second is integrity verification. The fact that Mior can forge the signed attestation report means that a mechanism intended to demonstrate that the VM is running the expected software can, in that scenario, lie. For an SMB that uses remote attestation as a compliance control or to validate environments, it is worth knowing that the mechanism has limits when the host is not trusted.
The third is vendor management. If your company contracts infrastructure from a small provider, a local integrator, or a private cloud set up by a third party, the level of control that third party has over the hypervisor and firmware is high. There the risk stops being theoretical and becomes dependent on the provider's maturity.
What we recommend from ForgeNEX
There is no need to panic or dismantle the infrastructure. What is advisable is to review assumptions and reinforce layers. These are the actions we propose to our clients:
- Classify your workloads. Distinguish which data needs protection from the provider itself (industrial secrets, regulated data) and which does not. Not all VMs require the same level.
- Ask your provider. Request information about their trust model, their access controls to the hypervisor, their firmware management, and their patching processes. A serious provider responds without discomfort.
- Do not rely only on memory encryption. Combine SEV-SNP with encryption at rest, your own key management (BYOK), and granular access controls. Defense in depth remains the best recipe.
- Review attestation. If you use it for compliance, document its limits and do not treat it as the sole proof of integrity.
- Keep firmware and the hypervisor up to date on any infrastructure of your own or in a private cloud. Attacks that depend on BIOS and hypervisor are mitigated, in part, with disciplined updates.
- Segment and monitor. Detect anomalous access to virtualization APIs and log who can touch the host layer.
This case is, at bottom, similar to others we have discussed on the blog: security is not only in the product you buy, but in how you configure it and in who holds the keys. We saw it with Check Point and the ESXi wipe, where access to the virtualization layer changes everything, and with UnrealIRCd, a reminder that flaws exploited in exposed services require a rapid response. The lesson here is the same: identify what you trust to whom and verify that this trust is justified.
Conclusion
Mior and Skyfall are not attacks you are going to suffer tomorrow morning from a phishing email. They are a wake-up call about the limits of confidential environments when the host is not trustworthy. For an SMB, the sensible response is not to abandon the cloud, but to choose well who you buy it from, demand transparency, and not base all security on a single technology. If you want to review how your infrastructure is set up and what assumptions should be questioned, at ForgeNEX we can help you carry out that analysis calmly.
Source: The Hacker News. Analysis and adaptation: ForgeNEX.
Keep reading
- Check Point unpatched: TSGA backdoor and ESXi wiping15 November 2026
- UnrealIRCd: exploited flaw allows command injection on the server15 November 2026
- Deepfakes and crypto: the North Korean fraud that deceives via video call13 November 2026
- SCKit exploits ScreenConnect: patch now or expose your remote support13 November 2026
- Critical Apache HTTP/2: unauthenticated remote code execution12 November 2026