Category Archives: Potential Risk of CVE

CVE-2025-33228: About NVIDIA Nsight Systems (23rd Jan 2026)

Preface: Nsight Systems is a tool for developers who need to understand the big picture of application execution on heterogeneous systems, especially in scenarios involving data transfer bottlenecks between the CPU and GPU or scaling across multiple nodes.

Background: In NVIDIA Nsight Systems, process_nsys_rep_cli[.]py is an internal Python script used primarily for post-processing and report generation from raw profiling data. 

While users typically interact with the nsys command-line tool, this script is invoked behind the scenes during the following operations.

Why This Might Connect to Nsight Systems?

Nsight Systems allows exporting [.]nsys-rep files and then processing them with scripts like process_nsys_rep_cli[.]py.

If the CLI or scripts read commands or code from user-provided files without validation, it could lead to:

-Command injection (similar to os[.]system()).

-Code execution (similar to exec()).

The design flaw could be that Nsight Systems assumes [.]nsys-rep or related files are safe, but if an attacker crafts a malicious file and you run the processing script, it could execute harmful commands.

Vulnerability details: CVE-2025-33228 NVIDIA Nsight Systems contains a vulnerability in the gfx_hotspot recipe, where an attacker could cause an OS command injection by supplying a malicious string to the process_nsys_rep_cli[.]py script if the script is invoked manually. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, data tampering, denial of service, and information disclosure.

Official announcement: Please refer to the link for details –

https://nvidia.custhelp.com/app/answers/detail/a_id/5755

CVE-2025-33233 – About NVIDIA Merlin Transformers4Rec for all platform  22nd Jan 2026

Official Updated 01/20/2026

Preface: Data engineers perform seamless preprocessing, a foundational stage where they gather messy, raw data from diverse sources, clean it (handling missing values, outliers, inconsistencies), integrate disparate datasets, and transform it into a unified, structured format, making it ready and reliable for data scientists to perform advanced feature engineering (creating new, meaningful features) and ultimately build better machine learning models. This ensures a high-quality, consistent input, preventing “garbage in, garbage out” for the modeling phase.

Background: Transformers4Rec is pre-installed in the merlin-pytorch container available from the NVIDIA GPU Cloud (NGC) catalog. This container is part of the NVIDIA Merlin ecosystem and is specifically designed to support sequential and session-based recommendation tasks using PyTorch.

The workflow can show you where we speculated design weakness of CVE-2025-33233.

NVTabular for preprocessingPyTorch for trainingTriton for serving—means PyTorch is a critical component. If its loading function is insecure, Merlin’s container is exposed regardless of NVIDIA’s own code. The workflow can display the location of suspected design flaw CVE-2025-33233.

If Transformers4Rec internally uses torch.load (which is common for loading PyTorch models) and relies on weights_only=True for safety, then CVE-2025-32434 could be the root cause or at least a contributing factor.

NVIDIA might have classified it as a separate CVE because the exploit path involves their product’s integration with PyTorch, making it a product-level exposure rather than just a dependency issue.

Vulnerability details: CVE-2025-33233 NVIDIA Merlin Transformers4Rec for all platforms contains a vulnerability where an attacker could cause a code injection issue. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, information disclosure, and data tampering.

Official announcement: Please refer to the following link for details-

https://nvidia.custhelp.com/app/answers/detail/a_id/5761

The media reports in January 2026 were triggered by a security warning issued by Apple on December 16, 2025! (20th Jan 2026)

Preface: This article is scheduled for release on January 21, 2026. It may contain some key topics that will be helpful to developers. Therefore it release today.

The “spyware” mentioned in April 2025 was the Apple Company policy, but the attacks reported in January 2026 are live threats using newer exploits (CVE-2025-43529 and CVE-2025-14174). If you have not updated to iOS 26.2 or newer, your device is at high risk from these “extremely sophisticated” attacks.

Background:

ANGLE on iOS: Chrome on iOS uses WebKit as its rendering engine (due to Apple’s App Store policy), and ANGLE is integrated for WebGL translation.

WebKit Zero-Day Exploits: If attackers exploit WebKit vulnerabilities (like CVE-2025-43529) and combine them with ANGLE flaws (CVE-2025-14174), they can potentially achieve remote code execution or device compromise.

Shared Components: Safari and Chrome on iOS both rely on WebKit, so any ANGLE-related bug in WebKit’s GPU pipeline could propagate to iOS.

Cyber security focus – CVE-2025-43529

When the JIT compiler requests executable memory via jit_heap and writes machine code, bmalloc/libpas allocates memory regions. If an object or memory block is freed prematurely and then accessed, that’s where UAF occurs. Complex DOM/JS operations or JIT optimizations often trigger these conditions.

The CVE-2025-43529 is tied to a use-after-free in WebKit’s internal memory management (bmalloc/libpas) during complex DOM/JS operations.

Cyber security focus – CVE-2025-14174

A security issue was found in ANGLE regarding gl.deleteTexture().

  • Asynchronous Release Bottlenecks: A known behavior is that calling gl.deleteTexture() does not immediately free memory if the GPU is still processing draw calls that reference that texture. Memory is only reclaimed after the GPU finished execution.
  • Shader Translation Memory Retention: A discovered bug in ANGLE’s PoolAllocator revealed that it often moves single-page allocations to a free list rather than returning them to the OS, causing memory used during shader translation to persist until the compiler instance is destroyed.

The following detailed information explains the relationship between ANGLE and different web browsers.

-Google Chrome Full Support (v1.0 & v2.0) Uses ANGLE

-Mozilla Firefox Full Support (v1.0 & v2.0) Uses ANGLE (on Windows) or own implementation

-Apple Safari Full Support (v1.0 & v2.0) Uses ANGLE (since v15+)

-Microsoft Edge Full Support (v1.0 & v2.0) Uses ANGLE (Chromium-based)

These vulnerabilities were actively exploited in the wild, often in spyware campaigns targeting high-profile individuals.

Even if you’re not a high-risk target, updating was critical because these flaws allow remote code execution via a crafted webpage.

For more details on these matters, please see the link.

https://support.apple.com/en-us/100100

https://www.forbes.com/sites/zakdoffman/2026/01/12/apple-warning-hundreds-of-millions-of-iphones-must-now-restart

CVE-2025-13328: About multiplexer control commands from the ETSI TS 07.10 20th Jan 2026

Mitre, NVD Official Published: 2026-01-16

Preface: ETSI TS 07.10 defines the multiplexer protocol for GSM networks, and DLCI 0 (Data Link Connection Identifier 0) specifically refers to the control channel used to manage multiple virtual serial sessions (like data, fax, voice) over a single physical serial link or its wireless emulation (like Bluetooth’s RFCOMM) to a device, essentially acting as the “local serial line” manager for those simultaneous connections on a smartphone or modem.

Background: In 2026, firmware-level flooding attacks over RFCOMM (Radio Frequency Communication) channels primarily exploit the device’s inability to handle excessive signaling and control traffic, leading to resource exhaustion or firmware crashes. These attacks typically occur without the need for prior pairing or authentication.

  1. Assumed earbuds firmware is the vulnerable component because it acts as the Responder on DLCI 0 and processes ETSI TS 07.10 control commands (like TEST) without proper limits.
  2. The attacker (or smartphone) sends the flooding traffic, but it’s lightweight for the sender—just repeated TEST frames.
  3. The earbuds do all the heavy lifting: parsing, allocating buffers, and responding. This leads to resource exhaustion on the earbuds, not on the smartphone.

The earbuds will run out of resources (CPU/memory) due to excessive TEST frames, not the smartphone.

Vulnerability details: This vulnerability is caused by the firmware’s susceptibility to flooding attacks over RFCOMM channels. When an attacker floods the standard control channel (DLCI 0) with a high volume of legitimate TEST commands, the device’s processing queue is overwhelmed, leading to resource exhaustion and a firmware crash that forcibly terminates paired user connections. Other active data channels across the device’s RFCOMM implementation are also vulnerable to flooding via MSC (Modem Status Command) signaling frames, including both the standard HFP (Hands-Free Profile) channel and an undocumented Airoha auxiliary service channel.

Official announcement: Please refer to the link for details –https://www.tenable.com/cve/CVE-2025-13328

CVE-2025-29943: CPU-caused stack corruption issue caused by flipping an undocumented MSR bit (19th Jan 2026)

Initial publication: 2026-01-15

Preface: CVE-2025-29943 allows disabling AES-NI at runtime, forcing mbedTLS to fall back to a vulnerable software AES implementation. In SGX enclaves, this enables cache-timing attacks to recover AES keys.

Attack Prerequisites:

  • Privileged access inside the VM (root or equivalent).
  • Ability to manipulate CPU feature bits or configuration.

Background: An SGX enclave (Software Guard eXtensions) is

a secure, isolated area within a computer’s processor and memory, creating a Trusted Execution Environment (TEE) for running sensitive code and data, protected even from the OS, hypervisor, or BIOS.

A specific undocumented bit can disable AES-NI at runtime, forcing the AES portion of mbedTLS to “fall back” to a version vulnerable to caching attacks. Utilizing this “fallback” in an SGX enclave, it can fully restored the enclave to use the AES key. Therefore, the vulnerability allows an attacker to extract (recover) the existing secret key.

Ref: The stack pointer (SP) inside an SEV-SNP guest is a critical CPU register, protected by SEV-SNP’s memory encryption and integrity features, but recent attacks like StackWarp and CacheWarp exploit microarchitectural side channels (often involving hyperthreading or cache timing) to infer or manipulate the SP and other sensitive state, potentially bypassing SEV-SNP’s security by observing how memory/cache changes during execution, allowing attackers to hijack control or leak data.

Vulnerability details: CVE-2025-29943: Improper access control within AMD CPUs may allow an admin-privileged attacker to modify the configuration of the CPU pipeline, potentially resulting in the corruption of the stack pointer inside an SEV-SNP guest.

Remark:

-Privileged accounts are not immune to compromise (e.g., insider threats, privilege escalation).

-Cloud or virtualized environments often assume strong isolation, but if an attacker gains root/admin inside the VM, they can exploit this vulnerability.

-The vulnerability affects cryptographic integrity and confidentiality, which is critical for SGX and secure workloads.

Reminder: It’s not a remote exploit, but local privileged exploits are still considered medium-risk (4.6) because:

  • They can lead to full compromise of sensitive data.
  • They break the security guarantees of SGX and SEV environments.

Official announcement: Please refer to the link for details.

https://www.amd.com/en/resources/product-security/bulletin/amd-sb-3027.html

CVE-2025-33206 – About NVIDIA NSIGHT Graphics for Linux (16th Jan 2026)

Preface: On Linux, this typically manifests as:

  • Unvalidated library loading (similar to LD_PRELOAD abuse).
  • Potential exploitation via malicious shared objects ([.]so files) placed in search paths.

Attackers could leverage writable directories or environment variables (like LD_LIBRARY_PATH) to inject malicious code when Nsight starts profiling.

Background:

Nsight Compute – As an interactive kernel profiler for CUDA applications, Nsight Compute provides detailed performance metrics and API debugging via a user interface and command line tool. Nsight Compute also provides customizable and data-driven user interface and metric collection that can be extended with analysis scripts for post-processing results.

Nsight Graphics – This is a standalone development tool for debugging, profiling and analysing graphics applications. Nsight Graphics allows optimization of performance of Direct3D 11, Direct3D 12, DirectX Raytracing 1.1, OpenGL, Vulkan, and KHR Vulkan Ray Tracing Extension based applications.

Injection allows tools like NVIDIA Nsight Graphics to sit between your application and the graphics driver to intercept, record, and manipulate the stream of commands.

Don’t underestimate the value of independent developer workstations. Sometimes, these workstations may store product prototypes and notes. If this information is intellectual property, and the workstation is located within the development team’s network segment, a cyberattack could not only lead to information leaks but also cause financial losses!

Vulnerability details: CVE-2025-33206 – NVIDIA NSIGHT Graphics for Linux contains a vulnerability where an attacker could cause command injection. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, data tampering, and denial of service.

Remark: If Nsight runs with elevated permissions (common for GPU profiling), exploitation can lead to:

  • Arbitrary code execution.
  • Escalation of privileges.
  • Data tampering or denial of service.

Official announcement: Please refer to the link for details –

https://nvidia.custhelp.com/app/answers/detail/a_id/5738

CVE-2025-68493: Apache Struts recently released a security advisory recommending that you check your code and upgrade. Otherwise, the new version of Struts will refuse to execute any existing code that contains security vulnerabilities. (14th Jan 2026)

Preface: Many large, enterprise-level companies across various industries use or have used the open-source Apache Struts framework for building Java web applications. Companies using this framework must ensure they are on a currently supported version and apply security patches immediately.

Background: The Model-View-Controller (MVC) is a software architectural pattern used in Java and other programming languages to separate an application’s logic into three interconnected core components: the Model (data and business logic), the View (user interface), and the Controller (input handling and coordination).

This separation of concerns makes applications easier to manage, test, and scale by allowing developers to modify one part of the application without significantly affecting the others.

Apache Struts is a framework for building enterprise-level Java apps. It follows the Model-View-Controller (MVC) pattern. Struts helps maintain a clean code structure and scalability. The framework supports internationalization, making it ideal for multi-language apps. It integrates easily with other Java EE components like EJBs and JMS.

Vulnerability details: Missing XML Validation vulnerability in Apache Struts, Apache Struts.This CVE (based on its pattern and recent advisories) likely relates to remote code execution (RCE) or OGNL injection triggered by unsafe configurations or certain result types. If your configuration:

  • Allows user-controlled parameters to influence OGNL expressions.
  • Uses developer mode in production.
  • Has wildcard mappings or dynamic method calls.

Official announcement: Please refer to the link for details –

https://www.tenable.com/cve/CVE-2025-68493

This issue affects Apache Struts: from 2.0.0 before 2.2.1; Apache Struts: from 2.2.1 through 6.1.0.

*Users are recommended to upgrade to version 6.1.1, which fixes the issue.

CVE-2025-47393: In Qualcomm-specified products, memory corruption occurs when the core driver accesses resources. 13th Jan-2026

Qualcomm – Official announcement: 1st Jan 2026

Preface: The Qualcomm Snapdragon Ride platform is used to develop advanced driver assistance systems (ADAS) and autonomous driving (AD) for vehicles. It combines powerful hardware (SoCs containing AI, GPUs, and vision engines) and software (SDKs, cloud tools) to support a wide range of functions from basic safety features to advanced autonomous driving. It allows for the integration of digital cockpit, ADAS, and AD functions on the same hardware and supports over-the-air (OTA) updates for continuous improvement.

Qualcomm SA9000P is a high-performance automotive-grade System-on-Chip (SoC) from Qualcomm’s Snapdragon Ride platform, part of a 5nm compute platform for advanced driver-assistance systems (ADAS) and autonomous driving, designed to compete with NVIDIA and Intel Mobileye, often paired with the SA8540P, enabling powerful in-car computing for future connected and self-driving vehicles.

Background: In the context of Qualcomm’s software ecosystem and the Linux kernel, _count_phandle_with_args() is typically a low-level helper or a variant of the standard DeviceTree (DT) API used to determine the number of phandle entries in a specific property.

While the internal underscore-prefixed version (_count_phandle_with_args) is often used within kernel core code (like drivers/of/base.c), it is most commonly accessed by Qualcomm drivers via the public wrapper: of_count_phandle_with_args()

Therefore, developers are advised to use `of_count_phandle_with_args()` to verify array indices.

Usage in Qualcomm Drivers: Qualcomm’s MSM (Mobile Station Modem) kernel and downstream drivers use this to dynamically determine how many resources (like regulator handles or clock inputs) are defined for a hardware block before allocating memory for them.

Vulnerability details:

CVE ID – CVE-2025-47393

Title – Improper Validation of Array Index in Automotive Linux OS

Description – Memory corruption when accessing resources in kernel driver.

Technology Area – Automotive Linux OS

Vulnerability Type – CWE-129 Improper Validation of Array Index

Official announcement: Please refer to the link for details –

https://docs.qualcomm.com/securitybulletin/january-2026-bulletin.html

CVE-2025-47345: About Qualcomm – Reusing a Nonce, Key Pair in Encryption in Automotive Platform (12th Jan 2026)

Official published: 5th JAN 2026

Preface: When used in an automotive context, the Snapdragon 8 Gen 3 Mobile Platform—and its dedicated automotive counterparts—utilize a High-Level Operating System (HLOS).

While the “Mobile Platform” is a consumer-grade chip typically found in smartphones, some automakers have integrated it directly into vehicles. For purpose-built automotive solutions, Qualcomm offers the Snapdragon Cockpit Elite and Snapdragon Ride Elite, which share the same underlying architecture (Oryon CPU cores) as the mobile 8-series.

Background: The Snapdragon 8 Gen 3 Mobile Platform redefines mobile connectivity with advancements in speed, reliability, and future-proofing. At its core is the Snapdragon X75 5G Modem-RF System, which delivers high performance, including up to 10Gbps downlink and 3.5Gbps uplink speeds.

While the High-Level Operating System (HLOS) allocates a non-secure buffer to communicate with the Trusted Execution Environment (TEE), this buffer is only used for passing encrypted payloads, commands, or non-sensitive handshake parameters.

The Automotive smart cockpit is an advanced integrated digital environment that uses artificial intelligence, sensors and connectivity to unify the driver interface, infotainment and vehicle controls, transforming the cabin into a personalized interactive space for driving, entertainment and productivity.

HDCP is used in smart cockpits to protect high-resolution digital video and audio content transmitted across internal digital interfaces like HDMI, DisplayPort, and automotive-specific links such as APIX (Automotive Pixel Link ) or GMSL (Gigabit Multimedia Serial Link).

HDCP in Automotive Platforms (HDCP) is a link protection protocol, and its security depends on proper key management and session handling inside the TEE.

  • The session keys are stored securely in TEE and referenced by session ID.
  • There’s no indication that the same nonce or key pair is reused across sessions or encryption operations.
  • The non-secure buffer is freed after the session ends, which is good practice.

Vulnerability details: Cryptographic issue may occur while encrypting license data. The potential vulnerability (CVE-2025-47345) in Qualcomm Snapdragon platforms arises from reusing a nonce or key pair during encryption, violating cryptographic best practices (CWE-323). This issue is not an application-level flaw but a platform design decision.

As a matter of fact, modern threat models consider:

•         TEE compromise or privilege escalation as realistic attack vectors.

•         Static cryptographic material as a critical weakness, enabling replay attacks, impersonation, or content decryption.

Official announcement: Please refer to the link for details –

https://nvd.nist.gov/vuln/detail/CVE-2025-47345

CVE-2025-47339 – Memory corruption while deinitializing a HDCP session -Use After Free in HLOS (9th Jan 2026)

Official Published: 01/05/2026

Preface: In Qualcomm devices, the Host Operating System (HLOS), often Android/Linux, manages HDCP (High-bandwidth Digital Content Protection) sessions by interacting with dedicated hardware/firmware (DCP/ MediaLink/TrustZone) for key exchange and encryption, ensuring protected content (DRM) is output securely over HDMI/DisplayPort, with the HLOS kernel handling driver calls and security enforcement to prevent playback of protected media on non-compliant displays.

Background: For the HLOS (Normal World) to communicate with the Secure World, a small “shared memory” buffer must be initialized:

• Communication Buffers: The HLOS allocates non-secure memory to pass non-sensitive commands and status updates (e.g., “start session,” “query status”) to the TEE.

• Buffer Alignment: Systems often require specific alignment (typically 4KB page alignment) for these shared buffers to ensure they can be mapped into the TEE’s address space for processing.

When the app calls mediaDrm.closeSession(sessionId) – refer tp attached diagram, the Widevine DRM stack signals the TEE (TrustZone) to terminate the secure session.

The non-secure buffer allocated by HLOS for communication with the TEE is freed once the session ends.  Alignment requirements (e.g., 4KB) are relevant only during active mapping; after deinitialization, the memory is returned to the normal pool.

Related details:

  • The HDCP link is not persistent beyond the DRM session. Once the session is closed, the secure channel is dismantled.
  • If another app or playback starts later, the entire handshake process (including HDCP negotiation) will run again.

Vulnerability details: CVE-2025-47339 – Memory corruption while deinitializing a HDCP session – Use After Free in HLOS.

One of the possibilities – When the HLOS frees the non-secure buffer after session closure, any lingering references (e.g., in the TEE driver or asynchronous callbacks) can still access that memory. If the cleanup sequence doesn’t enforce strict ordering—such as ensuring all secure-world operations have completed before freeing the buffer—the freed memory could be reused by another process, leading to corruption.

Official announcement: Please refer to the link for details –

https://docs.qualcomm.com/securitybulletin/january-2026-bulletin.html