Graphics DDK CVE-2026-41155

Exploitation requires a local, low-privilege account on a host with Imagination Technologies GPU hardware running an affected Graphics DDK kernel module (versions 1.18 RTM, 23.2 RTM, 24.2 RTM, 25.1-25.3 RTM, or 26.1 RTM). The attacker must be able to invoke GPU kernel module interfaces to allocate or manipulate secure GPU memory - typically gated by access to the GPU device node, which on default Linux systems requires membership in the 'video' or 'render' group, or equivalent. The attack is limited to systems where multiple distinct users or processes are actively using secure GPU contexts simultaneously; single-process or single-user GPU deployments have no exploitable peer process to target. Remote exploitation is not possible (AV:L confirmed by CVSS vector). No non-default configuration setting is required beyond the presence of an affected DDK version.

The CVSS 3.1 base score of 5.5 (Medium) reflects local access (AV:L), low complexity (AC:L), low privilege (PR:L), no user interaction (UI:N), unchanged scope, and high availability impact with no confidentiality or integrity impact. However, the vendor-assigned 'Information Disclosure' tag and the description's explicit mention of data passing between secure processes suggest a confidentiality dimension not reflected in the C:N metric - this is a notable inconsistency that defenders should verify with Imagination Technologies. EPSS at 0.02% (5th percentile) and SSVC exploitation status 'none' with automatable 'no' and technical impact 'partial' collectively indicate this is low real-world exploitation priority at present. The vulnerability is not automatable (SSVC) and requires local access, constraining its blast radius to multi-tenant or cloud GPU environments where untrusted co-tenant workloads share GPU hardware. In single-user embedded or consumer device deployments the risk is further reduced. No public exploit or KEV listing exists, confirming this remains a theoretical-to-low-probability threat.

An attacker with a low-privilege local account on a system running an affected Graphics DDK version (e.g., a co-tenant in a cloud GPU instance) launches a malicious GPU workload that deliberately allocates shared secure memory segments. By inspecting the allocations, the attacker reads sensitive data belonging to a peer secure GPU process (such as inference outputs or framebuffer contents) or deliberately corrupts shared state to crash the peer process and force a GPU hardware recovery, denying GPU service to other users. No public exploit code exists at time of analysis.