Windows Hyper-V CVE-2026-45641

Attacker must already have low-privilege local code execution on a system with the Hyper-V role/feature enabled - either as a user inside a Hyper-V guest VM or as a logged-in low-privilege user on the host - and must be able to invoke the vulnerable hypervisor code path (hypercalls or virtual-device interactions). Limiting factors: AV:L precludes remote network-only attacks, PR:L requires prior authenticated access, UI:N means no user interaction is needed once the attacker is present, and only x64 builds with Hyper-V enabled are affected - systems without virtualization (Hyper-V role uninstalled, Windows Sandbox/WSL2/Credential Guard disabled) are not in scope.

The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/C:H/I:H/A:H, 7.8 High) describes a low-privilege local attacker achieving total CIA impact, which is consistent with a guest-to-host or local privilege escalation scenario via Hyper-V. EPSS at 0.15% (35th percentile) and SSVC 'Exploitation: none, Automatable: no, Technical Impact: total' indicate that while the worst-case outcome is severe (full system compromise), exploitation is non-trivial and not currently observed in the wild - this is not CISA KEV listed. The 'execute code locally' phrasing in the description is consistent with the AV:L vector; however, since Hyper-V is the attack surface, the realistic threat model is a malicious VM guest reading host memory, which raises priority for multi-tenant or untrusted-guest environments (e.g., Azure-style hosting, malware analysis sandboxes) above what raw EPSS suggests. Defenders running single-tenant workstations with Hyper-V disabled have materially lower exposure than virtualization hosts.

An attacker with low-privilege code execution inside a Hyper-V guest VM (or a local user on a Hyper-V-enabled host) issues crafted hypercalls or device I/O that triggers a type-confusion-driven out-of-bounds read in the hypervisor, leaking host kernel memory and pivoting to arbitrary code execution in the root partition. This effectively breaks VM isolation, allowing the attacker to compromise the host and any co-tenant VMs. No public exploit is identified at time of analysis, and SSVC marks exploitation as 'none' and not automatable.