Arm Mali GPU Drivers EUVD-2026-72802

Requires local code execution on a device running one of the affected Arm Mali GPU kernel driver versions (Bifrost r41p0-r51p0/r54p1-r54p2, Valhall r41p0-r55p0, or 5th Gen r41p0-r55p0). No elevated privileges are required - the vulnerability is exploitable by any user-space process with access to the GPU driver interface, including sandboxed Android applications. No network access, user interaction, or special configuration is required. Exploitation is limited to local information disclosure (reading freed memory); remote exploitation is not possible given AV:L. The availability of a GPU on the device and an affected driver version are the only prerequisites.

The vendor-provided CVSS 3.1 vector (AV:L/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N) scores this at 4.0 - a local, low-complexity attack with only limited confidentiality impact and no integrity or availability consequence. The PR:N rating with AV:L is notable: it suggests even sandboxed or minimally privileged processes can trigger the flaw, which is plausible for GPU driver interfaces exposed to user-space apps. EPSS at 0.14% (4th percentile) indicates very low exploitation probability in the near term. No KEV entry and no public exploit code have been identified. The 'Denial Of Service' tag in the metadata suggests DoS potential (kernel crash from UAF), which is not reflected in the vendor's CVSS A:N rating - this is a discrepancy worth noting. The practical threat on Android mobile devices is a malicious app reading freed kernel memory to extract sensitive data from prior GPU operations (e.g., frame buffer residues, prior process data). The attack complexity is genuinely low once local code execution exists.

A malicious Android application with no special permissions installs on a device using a Bifrost or Valhall GPU (e.g., a Samsung Exynos or MediaTek device). The app issues a sequence of valid GPU memory processing operations that trigger a use-after-free condition in the kernel driver, then reads the freed memory region to extract residual data - potentially including fragments of GPU frame buffers or kernel heap content from prior processes. No public exploit code exists at time of analysis.