Windows ReFS CVE-2026-50362

Exploitation requires the victim to mount or open an attacker-crafted ReFS volume (a malicious VHD/VHDX image, removable disk, or storage device with malformed ReFS metadata) on a host whose ReFS driver parses it - the UI:R metric confirms user interaction is mandatory and AV:L confirms the attacker must deliver the volume locally rather than reach it over a network. No authentication is required (PR:N), so any user who mounts the volume can trigger the flaw, but the attack is limited by the need for that mount action, local delivery of the media or image, and the presence of the vulnerable ReFS component on the listed Windows 10/11 and Server 2016-2025 builds. Systems that never mount untrusted ReFS volumes are not realistically reachable.

The CVSS 3.1 vector CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H yields 7.8 (High), driven by full confidentiality, integrity, and availability impact but offset by a local attack vector and required user interaction. The AV:L combined with UI:R is the key limiter: the attacker cannot reach this over a network and depends on a user mounting or opening the crafted ReFS volume, which is why PR:N (no authentication) does not translate into trivial mass exploitation. No EPSS score, SSVC decision, POC, or KEV listing was provided, so real-world exploitation pressure appears low-to-moderate rather than urgent - this is a credible code-execution primitive but not a wormable network bug. The PR:N alongside a local vector reflects the removable-media / mount-a-volume exploitation model rather than authenticated local access, so defenders should treat it as a targeted or social-engineering-dependent threat pending any EPSS or KEV signal.

An attacker crafts a malicious ReFS-formatted volume - for example a VHDX disk image or a USB drive with corrupted ReFS metadata - and delivers it to a victim via email, download, or physical media. When the user mounts or opens the volume, the ReFS driver parses the malformed metadata, triggers the heap overflow in kernel memory, and executes attacker-controlled code. No public exploit is identified at time of analysis, and the local attack vector plus required user interaction mean the attack hinges on convincing the victim to mount the crafted volume.