EUVD-2026-24628
GHSA-mqr4-9x5m-973r
Severity by source
AV:L/AC:L/PR:N/UI:R/S:U/C:L/I:L/A:H
Primary rating from NVD · only source for this CVE.
CVSS VectorNVD
Lifecycle Timeline
4DescriptionCVE.org
Improper validation of STRING tensor offsets could allows malformed string metadata to trigger out of bounds access during constant tensor import in Samsung Open Source ONE Affected version is prior to commit 1.30.0.
AnalysisAI
Improper validation of STRING tensor offsets in Samsung Open Source ONE prior to commit 1.30.0 allows local attackers with user interaction to trigger out-of-bounds memory access during constant tensor import, potentially causing information disclosure, data modification, or denial of service. The vulnerability affects the tensor metadata parsing logic when processing malformed string tensor definitions.
Technical ContextAI
Samsung Open Source ONE is a neural network compiler and optimization framework. The vulnerability resides in the constant tensor import functionality, specifically in the validation logic for STRING tensor offset metadata. STRING tensors in neural network frameworks require careful offset tracking to prevent buffer overruns. The improper validation of these offsets (CWE-1284: Improper Validation of Specified Quantity in Input) allows an attacker to craft malformed tensor metadata that bypasses bounds checks, leading to out-of-bounds read or write access to memory regions adjacent to the tensor data structure. This occurs during the import phase when the framework parses and instantiates constant tensors from model files or serialized formats.
RemediationAI
Apply the fix available in commit 1.30.0 or later of the Samsung Open Source ONE repository. Users should upgrade to version 1.30.0 or newer once released as a stable version. The fix is available at https://github.com/Samsung/ONE/pull/16481. Until patching is possible, implement input validation on neural network model files before import by using a model verification tool or sandboxing the model loading process in an isolated environment. Restrict file system permissions to prevent unauthorized modification of model files that could introduce malformed tensor metadata. Disable automatic model loading from untrusted sources and require explicit user review before importing neural network models.
This vulnerability allows remote attackers to execute arbitrary code on affected installations of VEEAM One Agent 9.5.4.
This vulnerability allows remote attackers to execute arbitrary code on affected installations of VEEAM One Agent 9.5.4.
Incorrect access control in the component /api/user/manager of One v1.0 allows attackers to access sensitive information
Remote unauthenticated code execution in Veeam ONE (CPE cpe:2.3:a:veeam:one) allows attackers to run arbitrary code on t
A vulnerability in Veeam ONE allows an unauthenticated user to gain information about the SQL server connection Veeam ON
Authentication coercion in Veeam ONE lets an unauthenticated network attacker force the Veeam ONE service account to per
A vulnerability that allows an attacker in possession of the Veeam ONE Agent service account credentials to perform remo
An improper access control vulnerability allows low-privileged users to execute code with Administrator privileges remot
Unauthenticated arbitrary file read in Veeam ONE exposes sensitive server-side files to remote attackers without requiri
Remote code execution in Veeam ONE, the enterprise backup monitoring and analytics platform, is achievable by high-privi
SQL injection in Veeam ONE ≤ 13.0.2 allows any authenticated low-privileged user to extract database contents over the n
NTLM credential capture in Veeam ONE exposes the Reporter service account to any low-privileged authenticated user over
Same technique Information Disclosure
View allShare
External POC / Exploit Code
Leaving vuln.today