Bifrost Gpu Userspace Driver CVE-2025-1246

The CVSS 3.1 score of 7.8 with vector AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H indicates: Attack Vector (Local) - requires local system access; Attack Complexity (Low) - exploitation is straightforward once access is obtained; Privileges Required (Low) - a non-privileged user can exploit this; User Interaction (None) - no user interaction needed; Scope (Unchanged) - impact is limited to the vulnerable component; and all three impact metrics (Confidentiality, Integrity, Availability) are High. The primary risk signal is the ability to exploit this through web APIs (WebGL/WebGPU), which could potentially lower the practical attack vector if sandbox escapes are chained. However, without confirmed KEV/CISA status, active exploitation data, or published POC, the real-world exploitation probability cannot be determined from provided data. The Local attack vector and Low complexity suggest this is a legitimate concern for multi-user systems and containerized environments. The lack of stated patch availability in the provided references is a significant concern indicator—if patches are not yet released for affected versions, this represents elevated risk during the patching window.

An unprivileged local user on a Linux system or multi-user Android environment could craft a WebGL shader or WebGPU compute shader that triggers out-of-bounds buffer access in the GPU driver's memory management routines. By carefully constructing GPU operations with malformed buffer descriptors, the attacker could read sensitive data from adjacent GPU memory (information disclosure), corrupt GPU memory structures to gain elevated privileges (privilege escalation), or crash the GPU driver (denial of service). On a web-accessible system, a malicious website could embed WebGPU code that triggers the vulnerability when visited by a user, potentially exfiltrating GPU-cached data without the user's knowledge. On mobile devices, a malicious app with GPU access could exploit this to read memory belonging to other applications or the system, bypassing GPU memory isolation guarantees. While the attack vector is local, the ability to trigger via WebGL/WebGPU significantly increases accessibility without requiring native code execution or special privileges.