Back Igorslab.De AMD Confirms Radeon Driver Vulnerability: Manipulated WebGPU Shaders Can Crash Applications
A specially crafted website causing problems with a graphics driver is hardly among the features users want from modern GPU acceleration. This is precisely the attack vector described by a security vulnerability that AMD published on November 10, 2026, under the identifier AMD-SB-4026. According to the manufacturer, the CVE-2026-48479 vulnerability affects certain Radeon graphics drivers and can be triggered by specially crafted shader programs. Particularly relevant here is that the security researcher demonstrated a potential attack via a web browser’s WebGPU interface.
AMD confirms that the reported behavior can be reproduced and that the affected process may crash as a result. The manufacturer therefore describes it as a potential denial-of-service attack. AMD has so far been unable to demonstrate more far-reaching effects, such as successful data manipulation or the execution of someone else’s code. The vulnerability is rated 6.9 on the CVSS 4.0 scale, placing it in the medium-severity category. However, the discrepancy between the potential memory corruption described by the researcher and the consequences actually confirmed so far is noteworthy.
According to AMD’s security advisory, the vulnerability originates in the user-mode drivers for DirectX and Vulkan. Among other tasks, these driver components handle graphics command processing and prepare shader programs for execution on the GPU. Shaders are small programs that run on the graphics processor. They have long served purposes beyond calculating lighting and surfaces in computer games. Modern graphics APIs also enable general-purpose computations, such as processing images or large amounts of data. WebGPU now makes corresponding functionality available to web applications as well.
This creates an additional interface between content from the internet and the underlying graphics drivers. A browser does not simply process arbitrary GPU commands without checking them. Instead, shaders and other inputs are validated by various software components and translated into a form suitable for the graphics hardware in question. According to AMD, however, security researcher Michał Andryskowski from Imperial College London demonstrated that a specially crafted shader can trigger an invalid memory access in the graphics driver. The manufacturer describes the cause as the use of an uninitialized or invalid state within the driver. The CVE description assigns the vulnerability to category CWE-787. This refers to writes outside the intended memory bounds, known as out-of-bounds writes. Such errors can occur, for example, when software uses invalid states or incorrect bounds when accessing a memory region.
AMD does not explain in its public security advisory which specific internal driver function is responsible for CVE-2026-48479. Consequently, the information available so far does not allow a reliable determination of which specific shader instructions or driver paths trigger the error. The unusual aspect of the reported vulnerability is the attack vector described. According to the researcher, a malicious shader program does not necessarily have to be introduced by a locally installed application. Instead, he was able to trigger the problematic processing via a web browser’s WebGPU interface.
WebGPU was developed to give web applications modern access to graphics and compute capabilities. This can offer significant performance benefits for such applications, since computationally intensive tasks can be executed directly on the GPU. At the same time, this increases the importance of securely processing untrusted inputs in browsers, graphics APIs, and drivers. AMD rates the vulnerability with a CVSS 4.0 base score of 6.9. The published assessment takes into account a potential network-based attack that requires neither prior authentication nor user interaction. However, this does not mean that AMD has confirmed a complete attack chain with more far-reaching effects.
So far, the manufacturer has only been able to reproduce a crash of the affected process. AMD has not identified reliable exploitation to modify data or take over a system. It is also not known whether the vulnerability is already being actively exploited. The security advisory provides no evidence of this. The list of affected products published by AMD is more specific than the general term Radeon graphics drivers might initially suggest. It primarily lists older Ryzen processors with integrated Radeon graphics, along with several embedded platforms. Affected consumer products include certain models from the Ryzen 4000 and Ryzen 5000 series, mobile Ryzen 7030 processors, and mobile Athlon 3000 models with Radeon graphics. AMD also lists several Ryzen Embedded families, for which security fixes are currently scheduled to be released only in December 2026.
This list does not allow us to infer that all Radeon graphics cards are affected. In particular, current dedicated Radeon RX models are not explicitly listed as affected products in AMD-SB-4026. However, this does not mean that all GPUs not listed are generally guaranteed to be secure. AMD merely states that products considered unaffected based on its current assessment do not appear in the corresponding overview. For embedded processors, the manufacturer also distinguishes between different variants of the V1000 family. Both the variants known by the codename Picasso and those known as Raven Ridge are explicitly listed.
There is an inconsistency in AMD’s documentation regarding the planned security fixes. The Adrenalin Edition 26.10.1 is listed as a mitigation in the Security Bulletin AMD-SB-4026 dated November 10, 2026. At the same time, the table states that the corresponding release is planned for November 2026. However, the official release notes for Adrenalin Edition 26.10.1 are already dated October 20, 2026. This driver version was therefore released before the security bulletin was published.
This timing discrepancy leaves the situation not yet fully clarified. For example, the security fix could be distributed through a specific driver variant or a separate support branch. However, the public security advisory provides no clear explanation for this. The regular release notes dated October 20 also do not explicitly list CVE-2026-48479 as a fixed vulnerability. Therefore, the version number 26.10.1 alone should not be taken as a blanket indication that all affected systems are already protected by the package originally released in October. What matters is which specific driver version AMD releases for the respective product family and whether it actually includes the required fix.
Manufacturer-specific driver packages may also play a role, especially on older Ryzen platforms. On laptops, graphics drivers are often customized and distributed by the respective device manufacturer. Users should therefore refer to AMD’s official security advisories and, where applicable, the support information provided by their laptop or motherboard manufacturer. For the Ryzen Embedded platforms, AMD has not yet specified version numbers for the planned security fixes. A release in December 2026 is currently planned only for the R1000, R2000, V1000, and V2000 series.
This is particularly relevant because embedded processors are often used in devices that operate largely unchanged for long periods. These include industrial controllers, specialized computer platforms, and custom systems. Whether the WebGPU attack vector described can actually be reached on such devices depends heavily on their software configuration. A system without a browser and without processing such shader content does not automatically have the same attack surface as an ordinary desktop PC. However, a vulnerable driver component may also be accessed through other applications if they reach the faulty processing path. It has not yet been established whether such alternative exploitation paths are practical for CVE-2026-48479.
For operators of such systems, this means that the actual risk must be assessed based on the installed software, accessible interfaces, and planned security fixes. In its Security Bulletin, AMD recommends generally exercising caution with unknown links and untrusted content. This is a reasonable general precaution, but it does not replace a technical fix for the vulnerable driver component.
For affected systems, the first priority is to determine which graphics driver version is actually installed and which updated packages AMD or the respective device manufacturer provides. On browsers that support WebGPU, restricting this feature under certain conditions can reduce the described attack surface. Whether this is necessary or advisable in a particular case depends on how the system is used and which security updates are available. However, the published advisory does not support a blanket recommendation to disable WebGPU on all Radeon systems. The list of affected products is too specific for that, and the damage confirmed so far is limited to a possible process crash.
Until clearly documented security fixes are available, users should follow the official driver and security advisories. For embedded systems, coordination with the respective device manufacturer is advisable, since drivers and firmware there are often part of a specially customized system configuration.
What I find most interesting this security advisory is the attack vector described. WebGPU gives web applications significantly more powerful ways to use modern graphics hardware. At the same time, this case shows that the quality of the underlying drivers remains a key prerequisite for running such applications securely. Based on the information available so far, the effects confirmed by AMD are limited. A reproducible process crash is frustrating and can certainly be problematic in certain environments, but it does not justify claims of a potential complete system takeover. There is no evidence to support that.
I consider the information the security fixes more concerning. If a Security Bulletin in November lists a driver version as a planned mitigation when its regular release notes date back to October, AMD should clearly explain this connection. Especially with security updates, users need to be able to tell whether their system is actually protected.
The vulnerability itself currently appears manageable. However, the somewhat contradictory documentation makes it unnecessarily difficult for affected users to determine whether they have the correct update level.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
