Windows DWM CVE-2026-34336

Requires local, authenticated access to an affected Windows or Windows Server host: the CVSS vector AV:L/PR:L/UI:N means the attacker must already be able to execute code as a low-privileged local user, with no user interaction needed. There is no network vector and no remote path - the attacker cannot reach this over the wire. Exploitation targets the DWM Core Library, which is present by default in the graphical Windows stack, so no special feature toggle is required; however, Server Core installations are also listed as affected. Limiting factors that reduce realistic risk: the need for prior local access and valid credentials, the manual (non-automatable) nature per CISA SSVC, and the absence of any public exploit.

The signals are consistent and point to a moderate, non-urgent priority. The CVSS vector (AV:L/AC:L/PR:L/UI:N) requires local access and low privileges but no user interaction, yielding 7.8 (High); however EPSS is only 0.04% (13th percentile), there is no KEV listing, and CISA SSVC scores Exploitation: none and Automatable: no - so real-world exploitation likelihood is low and would be manual and targeted, not wormable. SSVC Technical Impact: total aligns with Microsoft's C:H/I:H/A:H rating. One notable inconsistency: the description says only 'disclose information locally,' yet the CVSS marks full integrity and availability impact - this discrepancy should be verified against Microsoft's MSRC advisory, as a pure buffer over-read (CWE-126) normally yields confidentiality impact rather than I:H/A:H. Net: patch on normal cadence rather than emergency, prioritizing multi-user and terminal-server hosts where local access is broader.

A low-privileged user who can already run code on a shared Windows or terminal server host invokes the vulnerable DWM Core Library code path and triggers the buffer over-read, causing it to return memory beyond the intended buffer. The attacker harvests the leaked bytes - potentially kernel or process pointers and other sensitive data - to bypass ASLR or feed a follow-on local privilege-escalation exploit. No public proof-of-concept is identified at time of analysis, and the attack is manual (SSVC Automatable: no).