Executive brief
OneFlow is an open-source deep learning framework used for building and deploying artificial intelligence models. A vulnerability in its CUDA-based boolean tensor component allows an attacker to crash the application by providing specifically crafted, invalid input values. This results in a denial-of-service (DoS) condition, potentially disrupting AI model training or inference services.
Technical details
A vulnerability classified as Uncontrolled Resource Consumption (CWE-400) exists in OneFlow v0.9.0 within the flow.cuda.BoolTensor component. The issue is triggered when the API is called with invalid values, such as extremely large integers or negative numbers (e.g., -8353862602220610428), which exceed the valid range for boolean types. This causes the underlying CUDA implementation to abort, resulting in a core dump and immediate process termination. The attack can be executed remotely if the application exposes tensor creation to user-supplied data without prior validation. As of the advisory, the issue was demonstrated on Ubuntu 22.04 with CUDA 12.2.
Affected products
- OneFlow-Inc OneFlow 0.9.0
Timeline
- 2025-06-30: disclosed: Issue reported on GitHub repository
- 2026-01-28: advisory: CVE published to NVD