Imagination Graphics DDK CVE-2026-41158

Attacker must already be able to execute code as an unprivileged local user on a device running an affected Imagination Graphics DDK (25.1 RTM through 25.3 RTM, or 26.1 RTM) and must have access to the PowerVR GPU device node to issue GPU system calls - typically granted to any normal Android app or any local user account on Linux/embedded systems with the driver loaded. No special non-default configuration is required beyond the driver being installed and the GPU exposed to userspace, and no user interaction is needed once code runs (UI:N). Exploitation is limited to systems actually shipping the affected Imagination DDK (PowerVR-based Android handsets, automotive, set-top, and embedded SoCs); devices using Mali, Adreno, or other GPU stacks are not in scope, and environments that deny GPU device access to untrusted processes via SELinux/MAC effectively block the primitive.

The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) correctly characterizes this as a local, low-complexity, low-privilege flaw with high impact across the CIA triad - a classic local privilege escalation primitive. EPSS is extremely low at 0.02% (5th percentile) and the issue is not on CISA KEV, with no public exploit identified at time of analysis, so mass exploitation is unlikely in the immediate term. However, raw EPSS understates risk here: GPU UAF bugs in mobile drivers (Mali, Adreno, PowerVR) have a strong track record of being weaponized for Android sandbox escapes and root, and Imagination GPUs ship in large fleets of Android and embedded devices, so targeted exploitation by commercial spyware or capable threat actors is plausible. Treat as a high-priority patch for any Android/embedded fleet shipping PowerVR GPUs, while accepting that opportunistic mass exploitation is not the dominant scenario today.

An attacker convinces a victim to install a malicious app (or compromises a low-privilege process on an embedded device with a PowerVR GPU), then issues crafted GPU system calls that allocate and free physical pages without deferred-free, while keeping GPU page-table entries pointing at those addresses. As the kernel reuses the freed pages for sensitive structures (credentials, page tables, kernel code), the attacker uses the GPU as a DMA primitive to read or overwrite them, escalating to root/kernel privileges. No public exploit is identified at this time, but the primitive is consistent with prior weaponized GPU UAFs on mobile platforms.