The Digital Arsenal | Research-based defensive guide | November 21, 2026. CVE-2026-99312 affects certain Linux KVM hypervisors on AMD processors when nested virtualization is active. DigitalOcean has validated a research demonstration that read host physical memory through a guest and described a separate controlled write demonstration. This checklist explains how an authorized administrator can determine whether their host environment is exposed, without reproducing the cross-tenant technique.
The vulnerable component is the KVM host kernel, not an arbitrary web application or every guest operating system. A guest user cannot fix an affected cloud provider’s hypervisor by upgrading software inside the guest. The first question is therefore: who operates the underlying host, and does it use nested virtualization?
Step 1: Confirm your responsibility and architecture
- Identify which systems you operate as physical hypervisors versus virtual machines consumed from a provider.
- Record CPU vendor, operating-system build, virtualization stack and whether the workload deliberately runs virtual machines inside virtual machines.
- Inventory hosted tenants and workloads to understand the potential consequences of a kernel maintenance change.
- Ask your infrastructure provider for a documented remediation statement where the hypervisor is outside your control.
- Do not launch guest-to-host proof-of-concept code on production infrastructure or other users’ systems merely to decide whether an update is needed.
Step 2: Inspect the host’s installed state
On an authorized Linux host, administrators can begin by documenting the output of uname -r and examining the operating-system’s supported packages. For AMD virtualization, check whether the kvm_amd module is loaded and review its documented nested parameter using the distribution’s management interface or /sys/module/kvm_amd/parameters/nested where present. A missing or differently represented setting should be investigated through distribution documentation rather than guessed.
This is an inventory check, not an exploit test. A value suggesting nested virtualization is enabled does not alone prove that the installed kernel is vulnerable. Conversely, a default-deny policy or unused nested capability may reduce reachability but does not replace a vendor’s security assessment.
Step 3: Match the patch to your distribution
The Linux kernel security announcement names upstream fixes in 6.12.118, 6.18.64 and 7.3.5. Distribution vendors such as Ubuntu and Red Hat may carry security backports under their own package versions and track affected package streams independently. Compare exact host packages, vendor errata and patch status: a simple comparison against the upstream number can wrongly label a backported kernel vulnerable or a custom kernel safe.
DigitalOcean’s November 19 disclosure says its provider-managed host fleet has been addressed with live patching. Its customers do not need to update their guests solely for this provider’s remediation. That statement applies to DigitalOcean’s relevant infrastructure, not to other cloud suppliers or customer-managed KVM hosts.
Step 4: Contain safely where immediate upgrade is impossible
For an operator who confirms they have a vulnerable AMD KVM host, DigitalOcean identifies disabling nested virtualization as a mitigation until a fixed host kernel is available. This setting can break workloads that depend on nested VMs, so it should be evaluated with service owners and documented through change control. Do not unilaterally change a shared host’s module configuration in the middle of sensitive workloads.
Step 5: Confirm the remediation
- Record the update approved, package identifiers and implementation time.
- Confirm that the fixed host kernel or validated vendor live patch is running, not merely downloaded.
- If disabling nested virtualization, verify the actual effective setting after the relevant host maintenance and check dependent services.
- Review operational telemetry for errors or regressions and keep a recovery plan ready.
- Store the proof of patch completion without copying tenant memory, customer datasets or unrelated secrets.
A note on severity
DigitalOcean treated the demonstrated isolation bypass as critical for its environment, whereas Red Hat’s published overall rating is moderate because exploitation requires specific nested AMD conditions and privileged guest access. Those different perspectives are not necessarily contradictory. Severity must be tied to exposure, ability to reach the vulnerable state and the consequences of crossing a trust boundary.
Research basis
DigitalOcean first-party disclosure and mitigation; Linux kernel CVE bulletin; Red Hat affected-scope and product status; Ubuntu release-specific tracking; IPCRIME vulnerability reporting.
No hands-on cross-tenant testing was conducted by IPCRIME. This guide is for authorized, nonintrusive host-inventory and remediation workflows.