Samsung Open Source ONE EUVD-2026-24628

Exploitation requires the following specific conditions: (1) Samsung Open Source ONE installation on the target system, (2) a model file with malformed STRING tensor offset metadata in a format supported by ONE's constant tensor import (such as flatbuffers, ONNX, or Samsung's proprietary format), (3) local user privileges to execute ONE's model import tools or APIs, and (4) user interaction to load the crafted model file. Exploitation does NOT require network access, administrator privileges, or pre-existing authentication. The vulnerability is NOT triggered by default configurations alone; a malicious or specially crafted model file must be present and explicitly imported. Default operation with legitimate model files is safe.

CVSS 6.6 indicates moderate risk with local attack vector and low attack complexity. The user interaction requirement (UI:R) and local-only access vector (AV:L) significantly constrain real-world exploitability compared to network-based vulnerabilities. However, the impact vector (C:L/I:L/A:H) shows denial of service as a primary concern alongside confidentiality and integrity impacts. No EPSS score or KEV status provided; without active exploitation confirmation, this represents a localized, availability-focused vulnerability rather than a widespread threat. The requirement for user interaction to load a malicious model file suggests exploitation depends on social engineering or supply-chain compromise of neural network models. The absence of public exploit confirmation means this is primarily a theoretical risk requiring deliberate triggering by a local user.

An attacker crafts a malformed neural network model file with deliberately invalid STRING tensor offset metadata and distributes it through a model repository or social engineering. A local user loads the model using Samsung Open Source ONE's model import functionality, triggering the vulnerable tensor parsing code. The out-of-bounds access crashes the process (denial of service) or potentially leaks sensitive memory contents depending on access type and memory layout.