A critical flaw has been identified in the Kernel-based Virtual Machine (KVM) hypervisor on AMD processors, designated as CVE-2026-99312. This vulnerability enables guest virtual machines (VMs) to read host memory, potentially exposing sensitive information from other VMs on the same physical host.
The vulnerability specifically affects cloud service providers utilizing AMD hardware with KVM and nested virtualization enabled. This includes numerous cloud platforms where multiple tenants share infrastructure.
The exploitation of this vulnerability could lead to significant data breaches, compromising customer data stored in other guest VMs. This poses a severe risk to the integrity and confidentiality of multi-tenant cloud environments.
- Update to the latest patched kernel versions provided by your Linux distribution.
- Disable nested virtualization where not required for operations.
- Monitor for unusual memory access patterns within VMs.
- Review and restrict VM configurations to limit potential attack vectors.
Key Technical Findings
Memory access vulnerability in KVM on AMD processors.
AMD processors with KVM enabled; specific kernel versions not specified.
Exploited through running malicious code within a guest VM.
Execution occurs through code running inside the guest VM that leverages the vulnerability to read host memory.
Not specified in the source material.
Not applicable; vulnerability allows reading memory without privilege escalation.
Not specified in the source material.
Potentially may expose sensitive credentials from other VMs in memory.
Not applicable; the focus is on memory access within the same host.
Data can be exfiltrated through reading sensitive memory contents from other VMs.
High; potential for significant data breaches and loss of customer trust.
Technical Background
The vulnerability identified as CVE-2026-99312 is a significant flaw within the Kernel-based Virtual Machine (KVM) hypervisor, which is integrated into the Linux kernel. This flaw allows guest VMs operating on AMD processors to access host memory directly when nested virtualization is enabled, creating an avenue for unauthorized data access and potential data breaches. The hypervisor typically enforces strict isolation between VMs to maintain security; however, this vulnerability undermines that core function, making it easier for one VM to snoop on another’s memory.
This attack vector leverages the inherent trust that hypervisors place on guest VMs, which are expected to operate in isolation. The typical objectives of an attacker exploiting this vulnerability include obtaining sensitive information, such as cryptographic keys or personally identifiable information (PII) stored in other VMs. As this flaw affects only AMD processors with nested virtualization enabled, it presents a unique challenge for cloud service providers who must balance performance and security across multi-tenant environments.
Attack Chain Analysis
-
Initial Access
Activity The attacker deploys a malicious or compromised VM instance within the cloud environment.
Evidence Unusual VM configurations or deployments that deviate from standard practices.
Telemetry Monitoring VM creation logs can reveal unexpected instances being spun up.
Detection opportunity Implement alerts for rogue VM deployments or unexpected resource utilization in shared environments.
-
Execution
Activity The malicious VM executes code that exploits the memory access flaw within the KVM hypervisor.
Evidence Memory read operations targeting specific addresses that contain sensitive information.
Telemetry Unusual memory access patterns can be captured through hypervisor logs.
Detection opportunity Utilize monitoring tools to analyze hypervisor logs and identify aberrant memory access attempts.
Deep Technical Behavior Analysis
The exploitation of CVE-2026-99312 hinges on the ability of a guest VM to perform memory read operations that should be restricted by the hypervisor. In typical scenarios, hypervisors maintain strict isolation between guest VMs, preventing any unauthorized access to another VM’s memory space. However, due to this vulnerability, attackers can manipulate their VM’s execution context to bypass these protections. This could involve executing specially crafted instructions that coerce the hypervisor into exposing its internal memory state, leading to potential information leakage.
This behavior can manifest as increased IO operations directed towards specific memory addresses associated with other VMs. Potential indicators include abnormal patterns of memory allocation or deallocation requests that do not align with the expected operational profile of the environment. Continuous monitoring and anomaly detection become essential strategies for identifying such exploitation attempts before they culminate in data breaches.
Indicators of Compromise
Indicators of Behavior
| Behavioral Indicator | Description | Data Source | Confidence |
|---|---|---|---|
| Unusual Memory Access Patterns | Repeated attempts to read from restricted memory addresses indicating potential exploitation attempts. | Hypervisor logs, EDR telemetry | Potential |
Detection Engineering Guidance
index=edr (event_type='memory_access' AND target_address='*specific_address*')
search index=cloud_logs source='vm_creation' | stats count by account_id | where count > 'X'



