CVE-2026-40450: Samsung ONE Buffer Overflow Vulnerability

CVE-2026-40450 Overview

CVE-2026-40450 is an integer overflow vulnerability affecting Samsung Open Source ONE (On-device Neural Engine), a machine learning compiler and runtime framework. The vulnerability occurs in the output tensor copy size calculation, where an integer overflow can lead to incorrect copy lengths and subsequent memory corruption when processing oversized tensors.

Critical Impact

This integer overflow vulnerability can lead to memory corruption, potentially allowing attackers to crash applications or achieve unintended behavior through maliciously crafted tensor inputs.

Affected Products

• Samsung Open Source ONE versions prior to commit 1.30.0

Discovery Timeline

• 2026-04-22 - CVE CVE-2026-40450 published to NVD
• 2026-04-22 - Last updated in NVD database

Technical Details for CVE-2026-40450

Vulnerability Analysis

This vulnerability is classified under CWE-190 (Integer Overflow or Wraparound). The flaw exists in how Samsung ONE calculates the copy size for output tensors. When processing tensors with extremely large dimensions, the multiplication operations used to compute the total copy size can exceed the maximum value representable by the integer type, causing the value to wrap around to a much smaller number.

This incorrect size calculation results in an undersized memory copy operation, which can lead to:

• Buffer overflows when subsequent operations assume the full data was copied
• Memory corruption affecting adjacent memory regions
• Application crashes or denial of service conditions
• Potential for exploitation if an attacker can control the tensor dimensions

The local attack vector requires user interaction, meaning an attacker would need to convince a user to load a maliciously crafted model or tensor data.

Root Cause

The root cause is improper validation and handling of arithmetic operations when calculating tensor copy sizes. When tensor dimensions are multiplied together to determine the total byte count for memory operations, the calculation does not check for or prevent integer overflow conditions. This is a common issue in deep learning frameworks where tensor shapes can vary dramatically and involve large numerical values.

Attack Vector

The attack requires local access and user interaction. An attacker could craft a malicious neural network model or tensor input file with oversized dimensions designed to trigger the integer overflow during output tensor copy operations. When a user loads this malicious input into an application using the vulnerable ONE framework, the overflow occurs, leading to memory corruption.

The vulnerability could be exploited through:

• Maliciously crafted .tflite or other model files
• Applications that process user-supplied tensor data
• Model conversion utilities that handle untrusted input

Detection Methods for CVE-2026-40450

Indicators of Compromise

• Unexpected application crashes when loading or processing neural network models
• Memory corruption errors or segmentation faults in applications using Samsung ONE
• Anomalous tensor dimensions in processed model files that appear unusually large
• Log entries indicating memory allocation or copy failures in ONE-based applications

Detection Strategies

• Monitor for crashes in applications that utilize the Samsung ONE framework
• Implement input validation to detect suspiciously large tensor dimensions before processing
• Use memory sanitizers (AddressSanitizer, MemorySanitizer) during development and testing to catch overflow-related corruption
• Audit model files for abnormal dimension values that could trigger overflow conditions

Monitoring Recommendations

• Enable crash reporting and analysis for applications using Samsung ONE
• Log tensor dimension values during model loading to identify anomalous inputs
• Implement runtime bounds checking for tensor operations in production environments

How to Mitigate CVE-2026-40450

Immediate Actions Required

• Update Samsung Open Source ONE to version 1.30.0 or later
• Review and validate all tensor inputs before processing, especially from untrusted sources
• Implement input size limits for tensor dimensions in applications using ONE
• Consider sandboxing applications that process untrusted model files

Patch Information

Samsung has addressed this vulnerability in GitHub Pull Request #16481. The fix is included in version 1.30.0 and later. Organizations using Samsung ONE should update to the patched version immediately.

To update, pull the latest changes from the Samsung ONE repository:

git clone https://github.com/Samsung/ONE.git
cd ONE
git checkout 1.30.0

Workarounds

• Implement pre-processing validation that checks tensor dimensions before passing to ONE
• Add overflow detection logic to custom code that interfaces with ONE tensor operations
• Restrict processing to only trusted model files from verified sources
• Run ONE-based applications in sandboxed environments to limit the impact of potential exploitation